MISH Networking · Built by Tyler
I set up the equipment, build the software and monitor the inner workings of your business, so it fits only yours.
Let’s simplify yoursEvery call, email, text and file goes through one machine in your building. The model reads your own history, drafts the next step, and you approve it. The result is stored, so the next draft starts from more of your own work.
Setups run from a single NVIDIA RTX PRO 6000 Blackwell server, with 96 GB of memory on each card, up to rack-scale systems like NVIDIA Vera Rubin NVL72: 72 GPUs, fully NVLinked into one machine.
The RTX PRO 6000 Blackwell links over PCIe 5.0; NVLink is the rack-scale link in systems like Vera Rubin NVL72. NVIDIA, RTX, Blackwell, Vera Rubin and NVLink are trademarks of NVIDIA. Sized to each business.
The cards and servers you begin with move into a server rack and keep working while the next one is built beside them, so what you buy first is an investment, not a placeholder.
Type one question. The day lists itself. Handled items cross off, and what needs you is marked Ready for you.
The call is transcribed as it happens. The box files it to the right client, pulls out what was promised and adds the follow-up to tomorrow’s list.
Calls are transcribed only where recording is allowed. Some states require everyone on the call to agree.
Everything above lives on one machine in your office. Your phone reaches it over a private connection, so you can ask the same box from anywhere.
The routine work is done. What’s left are the two calls only you should make.
Ten fictional examples. Explore one.
Sample business. Designed by Tyler, built with Claude.
Prepared examples, not connected businesses. Choose a capability to see the workflow and who approves it.
I match the AI to your business. Local AI, on a machine in your building, does the work that touches your records. For everyday desk work with the names and personal details taken out, I can also put AI agents to work that run on the large frontier models. If one of those models is the better fit for a job, I'll tell you, and show you exactly how I'd automate it. Your team approves what matters.
The model and hosting arrangements are proposed choices for fictional workflows, not deployed systems or a default for your business.
MISH
Day or night, I'm your direct contact for support across every system I build.
Work directly with meTyler B. Mishoulam · Austin, TX
On-site nationwide.
Why it runs in your building
You can hand a model every book and every line of code ever written, and it still won’t know which step your team quietly skips, or how your best person reads a customer. That knowledge lives in your business. A local model learns it.
Every call, email, text and file goes through one machine in your building. The model reads your own history, drafts the next step, and you approve it. The result is stored, so the next draft starts from more of your own work.
Your model
Each pass around the loop leaves more of your own history behind. Better drafts free up time, that time goes into more work, and the work adds more history. A rented tool is built for everyone. Yours learns from you and your team’s daily input, and what it learns stays in the building.
During the day, everyone has their own piece of the pie. It’s called an instance. And at night, each instance goes back into the server. And the server learns. Gets better, improves. And finds the company’s patterns: what’s working right and what is not working efficiently.
From there, it narrows down the process to the best way of getting things done for the most optimal outcome.
Local AI, on a machine in your building, does the work that touches your records. For everyday desk work with the names and personal details taken out, I can also put AI agents to work that run on the large frontier models. If one of those models is the better fit for a job, I’ll tell you, and show you exactly how I’d automate it.
Your morning, from your side of the desk
Every call, email, text and letter flows through one chat box on a machine in your office. It learns from all of it, so when you ask what needs doing, nothing is missing.
Good morning. Overnight: 2 cancellation notices, 9 renewal offers, 1 certificate request, 4 voicemails. 3 things need you.
She asks
Tell me what I need to do today.
The box tells her
Done Texted the 2 clients whose policies cancel for non-payment. Reminders keep going until they pay.
Done Opened a policy change from last night’s voicemail: the Ruiz family is adding a 16-year-old driver.
Ready for you The certificate for Cedar Bridge Framing is filled in, and their policy allows the wording. Approve to send.
Ready for you One carrier paid 3 October commissions short. The letter asking for the difference is drafted.
Caller Ana Ruiz
“Hi, it’s Ana. Our son just turned sixteen and got his license, so we need him on the car.”
Filed to the Ruiz auto policy. Policy change opened: add a driver.
Follow-up set for tomorrow.
She asks
Tell me the top 5 closed sales and what areas they were in this past month.
The box tells her
Your top 5 closed sales this past month:
Good morning. 4 carriers answered overnight, 2 checks were issued, 1 deadline is this week.
She asks
Tell me what I need to do today.
The box tells her
Done The lender has endorsed the Nguyen check, step 2 of 3. Their next step is due Thursday, and they’ve been reminded.
Done The carrier’s reply on the Diaz claim is filed to the claim, and the next follow-up is on the calendar.
Ready for you The Brooks roof is finished. The request to release the held-back depreciation is ready to send.
Ready for you The carrier’s estimate on the Ortiz claim leaves out 6 items. The supplement is drafted with photos.
Caller Dale Hart, field adjuster
“We’ll include the north gutters and the fascia on the revised estimate.”
Saved to the Ortiz claim with the date, time and name.
Added to the supplement checklist.
She asks
Tell me the five best areas right now to market my next roofing campaign based on parameters that have worked in the past.
The box tells her
The five best right now, on what has worked for you: hail in the last 12 months, roofs over 15 years old and past jobs nearby.
Good morning. 3 offers came in overnight, 2 showings moved, and 1 inspection report is back.
She asks
Tell me what I need to do today.
The box tells her
Done Confirmed tomorrow’s 4 showings with the buyers’ agents. Reminders go out at 8 am.
Done Filed the Yarrow Close inspection report and flagged the 2 repairs it lists.
Ready for you The counteroffer on Wren Hollow is drafted. Approve to send.
Ready for you The price-change note for the Marlow Street sellers is ready for you to read.
Caller Priya Shah
“We’d love to see the Marlow Street house again this weekend and bring my parents.”
Filed to the Shah buyer file. Second showing noted.
Saturday showing added to tomorrow’s list.
She asks
What are the top five fastest selling listings that I’ve had this month? And tell me if there’s anything specific that I did, for them or for all of the listings that I’ve sold so quickly in the last year.
The box tells her
Your five fastest this month:
What you did: all five had a video walkthrough up before the first showing and went live on a Thursday. Those same two moves are on 9 of your 12 fastest sales this year.
Good morning. 3 patients’ insurance needs a look before they arrive. 11 voicemails overnight, 2 marked urgent.
She asks
Tell me what I need to do today.
The box tells her
Done Overnight voicemails are sorted: 2 urgent at the top, 4 refill requests in the doctor’s queue.
Done Coverage is checked for tomorrow’s patients. 3 have an issue to look at before they arrive.
Ready for you 6 patients were told they need treatment and never booked. Reminder texts are ready.
Ready for you Mia Chen’s specialist hasn’t sent a report back. The request is drafted.
Voicemail from Ben Ortiz
“I need to move my crown appointment on Thursday. Mornings work better.”
Two morning times are held.
A text with the options is ready.
She asks
Tell me the five patients I should call first this week, and why.
The box tells her
Call these five first:
Good morning. 2 court notices came in, 14 clients still owe tax documents, and 3 client calls from yesterday aren’t billed yet.
She asks
Tell me what I need to do today.
The box tells her
Done Conflict check on yesterday’s new caller, Daniel Park: no matches in past matters.
Done The Travis County notice is read, and the hearing date is on the calendar for you to confirm.
Ready for you Yesterday’s 3 client calls are now draft time entries. Approve to bill.
Ready for you 14 clients still owe tax forms. Reminders are ready, and the late K-1s are flagged.
Caller Rachel Kim
“The K-1 from the partnership should come Friday. I’ll send it as soon as I have it.”
Expected Friday.
Reminder set, and the return stays on hold until it arrives.
She asks
Hey, set a reminder for October 28th at 2:30, I have a meeting. Remind me twice, once 30 minutes before and once two hours before.
The box tells her
October
SMTWTFS12345678910111213141516171819202122232425262728293031Reminders set. Do you need any documents on this day?
Prepared scenario. Fictional names and messages. The example compresses time. No email, appointments or payments are sent from this page.
Overnight, the calls, voicemails, notices and letters land in the box before you sit down.
Type one question. The day lists itself. Handled items cross off, and what needs you is marked Ready for you.
Switching swaps the greeting, the notifications, the list and the call for that business.
The call is transcribed as it happens. The box files it to the right client, pulls out what was promised and adds the follow-up to tomorrow’s list.
Everything above lives on one machine in your office. Your phone reaches it over a private connection, so you can ask the same box from anywhere. Nothing in the workday slipped, because all of it went through the box.
Check on a project between appointments. Find the hold-up before it becomes tomorrow’s problem. Ask what needs your attention and request the next step, right from iMessage.
I connect supported tools to a private assistant workflow on the machine in your office. The goal is straightforward: useful answers and approved work, without opening every app yourself. You choose what to connect and what the system has permission to inspect.
See the projects, deadlines and team activity you’re authorized to review. Answers should point back to the permitted records and flag anything missing or uncertain. A draft is not a completed action, and a request is not proof that work ran.
Your access. Your rules. Your approval. I scope each setup around authenticated access, limited permissions, consent and audit records. Sending messages, spending money, changing access or deleting records requires your approval through the agreed controls, not just an incoming text.
Availability depends on supported integrations, device availability and the access you grant. Security and compliance requirements are assessed for your industry and workflows before activation. These are example requests, not a live connection to your business.
Explore a phone-first setupThe routine work is done. What’s left are the two calls only you should make.
Sample business. Designed by Tyler, built with Claude.
How it gets into your building
Paper stacks on every desk. That’s where the week goes.
AI-generated footage.
I sit with your team and map where the hours actually go.
Illustration.
Your emails, files and records, gathered on your own hardware.
Illustration.
A private model, trained on how your business actually works.
Illustration.
Checked against real work before anyone relies on it.
Illustration.
Measured on work it has never seen. Before vs after.
Illustration, not a measured result.
Installed in your building and wired to every desk. Your records stay there.
AI-generated footage and illustration.
At the end of the day the flow reverses. Everything your team put in runs back to the server, and it learns from it.
AI-generated footage and illustration.
I keep watching how it runs and fix what slips, before your team feels it.
AI-generated footage and illustration.
Every approved answer makes it better. Next month starts from everything this month taught it.
AI-generated footage and illustration.
Your website, marketing and daily operations should work together. I connect the path from being found to getting booked, delivering the work and earning the next visit.
Start with one useful improvement or a coordinated plan for the whole business. The free assessment covers your website, visibility and operations, including AI where it fits.
I improve your website and web presence, connect sales to operations, and automate repetitive tasks so your business has room to grow. I build on the systems you already run.
I design or improve your website around a clear offer, fast mobile pages and an easy way to call, book or buy. Booking, forms and your CRM connect so an inquiry has somewhere to go.
Measure: qualified inquiries and completed bookings, not just visits.
Search visibility, accurate local business listings, useful content and consistent social profiles. I connect campaign landing pages to source tracking so you can see which channels bring useful inquiries.
Measure: relevant search traffic, lead sources and cost per qualified inquiry where spend applies.
Lead intake, missed-call follow-up, scheduling, reminders, CRM updates, invoicing and reporting. I automate repeatable work across your existing tools, with clear exceptions and human approval for consequential actions.
Measure: response time, manual hours and corrections after review.
Follow-up, rebooking, customer retention and owner reporting. I connect the customer journey so you can see where inquiries stop, which work converts and where there is room to grow.
Measure: booking conversion, repeat business and margin after operating costs.
Build against a baseline, not a promise. I agree priorities, scope, costs and success measures with you before building. Search rankings, lead volume and revenue growth are not guaranteed.
Your business stays yours. Your team approves access, public content, campaigns and consequential actions. Approved routine messages can run automatically with consent and opt-out rules. Integration and platform limits are checked first.
MISH. Sample business. Designed by Tyler, built with Claude. HALE, Health, longevity & care
112 biomarkers, read overnight by a model that runs inside the clinic. Your physician reviews every finding before it reaches you.
112markers read overnight
5 flagged for your physician
flag-drift, then draft-care-plan, both on the local model in the clinic's building.Example diagnosis, prepared with Claude. Not a live call.
Prepared care-coordination example: eligibility and benefits, patient-consented carrier coordination, clinician review, appointment confirmation and pharmacy stock checks. No live patient services.
ApoB, HbA1c, hs-CRP, ferritin, vitamin D and the rest, read against your last six panels. Drift is flagged in amber.
Example monitor, not patient data.
Local model in the clinic's building · I set it up · Physician signs
flag-driftdraft-care-plan: three changes, one retestWhy local: health records stay inside the clinic. I write both skills at setup from the clinic's written protocols, not from member records.
MISH. Sample business. Designed by Tyler, built with Claude. Marketplays, Marketplaces & social networks
Create one listing, choose the right marketplaces and keep every buyer conversation together. See one item move from a prepared listing to an approved sale, without switching apps.
1 post → 6 example destinations → 1 inbox
write-listing and draft-reply on the local model. Delisting is a plain rule.Example diagnosis, prepared with Claude. Not a live call.
The buyer receives your approved pickup details. You both know what happens next.
Tested. Strap included.
Vintage 35mm film camera, tested, strap included.
Prepared example complete · sold for $140
One item.
Selected destinations.
Zero retyping.
Does the light meter still work?
Could you do $120 shipped?
Is the strap included?
I can pick it up today. $140?
Perfect. Send the pickup details and I’ll see you there.
90 minutes.From post to sale in this example. Not a measured average or a guarantee.
Prepared demonstration. No listings, messages, payments or marketplace connections are live.
A proposed connection map, not live integrations. Availability depends on each platform’s permission, account eligibility, product category and region. Platform fees and rules still apply. A camera does not belong on every marketplace.
Local model writes and drafts · Rules delist · You approve
write-listing. Suggests a price from recent sales nearby with find-comps.draft-reply. It never sends on its own.Buyer messages still need permission, limited access and an approved data-handling setup before a real connection is enabled.
Marketplace names are trademarks of their respective owners.
MISH. Sample business. Designed by Tyler, built with Claude. Plumbline Construction, Construction
Prepared example. A fictional company. Nothing is ordered or sent from this page.
read-contract and compare-suppliers on the local model.Example diagnosis, prepared with Claude. Not a live call.
Someone signs the contract. Everything after that happens on its own, and every step points back to that page.
The moment it's signed, the chain starts. Nobody has to pick up a phone.
The shingles go on an order straight from the contract's own numbers.
Another job next week uses the same shingle, so the order waits and joins it. Bulk price, automatically.
Delivery goes to one spot right beside the house, not a yard across town.
It checks each supplier for that shingle and finds the best price.
It looks up the forecast for the install date before anyone is told to show up.
Every installer is asked if they're ready for that date, and each answer comes back on the board.
The manufacturer is contacted to be on site that same day, with 6-inch gutter, because the contract says 6-inch.
Every order, message and date traces to one signed page. That's how you show the job is being done the way it was sold.
Rules you set handle the routine. read-contract and compare-suppliers run on a local model in the office. You decide what needs your OK before it goes out.
Build this for your companyRules code the order and the schedule · Local model reads the contract and compares suppliers · Office approves
MISH. Sample business. Designed by Tyler, built with Claude. Porchlight Tax Protest, Property tax protest
Prepared example. A fictional company. No number here is a real result, and nothing is filed from this page.
find-matching-homes and draft-protest on the local model.Example diagnosis, prepared with Claude. Not a live call.
Every spring the county mails a new value for your home. This firm checks it against the homes around you, and you approve every letter before it goes.
The county raised your home's value. The notice goes straight into your file.
The date is read right off the notice and starts counting down.
Every home around you, from the firm's own files.
Same size, same age, same street. Only the homes like yours turn on.
The house down the street sold for less. Each sale goes up beside your value.
The sales and your notice go into one packet, written the way this firm writes them.
You read it and you sign. Nothing goes in until you do.
Every client got a notice that same spring. Every one gets the same steps, from the same files.
Rules you set handle the dates. find-matching-homes and draft-protest run on a local model in the office. You approve every protest before it goes.
Build this for your companyRules watch the dates · Local model finds the matching homes and drafts the protest · The owner approves
MISH. Sample business. Designed by Tyler, built with Claude. Solstice, Energy & home solar
Prepared example. The homeowner approves reserve changes.
Home battery · solar
Fixed dispatch rules plan when the battery stores sunlight, charges, sells at peak rates or runs the house. This is one prepared example day.
draft-summary.Example diagnosis, prepared with Claude. Not a live call.
45%battery in example day
Prepared example · Human approval
draft-summary. The homeowner approves any change to the backup reserve.MISH. Sample business. Designed by Tyler, built with Claude. Lucky 7 Pawn, Pawn & resale
Guitars, gold, consoles and one-of-a-kind finds. Everyday value, with prices checked against similar sold items. The clerk checks condition before making an offer.
Guitars, jewelry, fur coats, PS4 and Xbox consoles, tablets and one-of-a-kind finds. Example inventory.
find-comps on the local model, then plain math for the range.Example diagnosis, prepared with Claude. Not a live call.
Prepared reminders, redemptions and reviewed forfeiture into retail inventory. No live messages or payments.
Ticket 2611
Sunburst · original case · serial logged
Hold period noted
Range ready · the clerk makes the offer
Local model finds similar sales · Plain math sets the range · The clerk makes the offer
find-comps.The local model is given the item's details, not the customer's.
MISH. Sample business. Designed by Tyler, built with Claude. The Ashbury, Real estate & property
read-intake, draft-reply and dispatch-vendor on the local model.Example diagnosis, prepared with Claude. Not a live call.
Prepared example: the tenant request and vendor arrival.
Unit 4B · 7:42 pm
There's water under the kitchen sink. Not a lot, but it keeps coming.
Approved by the manager
Example figures for a sample building.
Local model triages and drafts · Records stay in the property system · The manager approves
read-intake.draft-reply, and picks the on-call plumber with dispatch-vendor.MISH. Sample business. Designed by Tyler, built with Claude. Ridgeline, Home services & roofing
Prepared exampleInspection to estimate
Photos in, line items out, checked by an estimator before the homeowner sees a number.
explain-estimate on the local model writes the note.Example diagnosis, prepared with Claude. Not a live call.
Prepared booking workflow: document the roof, review the measurements, check the draft.
| Item | Qty |
|---|---|
| Tear-off, 1 layer | 32.4 SQ |
| Synthetic underlayment | 32.4 SQ |
| Architectural shingles (+10% waste) | 35.6 SQ |
| Starter strip | 212 LF |
| Ridge & hip cap | 64 LF |
| Drip edge | 212 LF |
| Pipe boots | 3 EA |
Draft · waiting for the estimator
Inspection completed. Estimate under review. Crew dates proposed only after approval.
Survey crew
Inspection documented
Estimator
Review required
Install crew
Schedule proposed
Example weather
Measured on the tablet · Local model writes the homeowner's note · The estimator approves
explain-estimate.The owner decides which job details the local model may read.
MISH. Sample business. Designed by Tyler, built with Claude. Scrapyard, Fan-favorite products
Example drop · Friday 10:00DAYS · HRS · MIN · SEC
draft-run.Example diagnosis, prepared with Claude. Not a live call.
Prepared example
Prepared example · Human approval
draft-run. Every number here is an example. The founders approve the run.MISH. Sample business. Designed by Tyler, built with Claude. Fairweather Law, Legal services
Prepared example. A fictional law office. Nothing is filed or sent from this page, and it is not legal advice.
find-similar-cases and summarize-case on the local model.Example diagnosis, prepared with Claude. Not a live call.
A new client walks in with a problem. The lawyer asks one question: have we seen this before?
She sets a letter on the desk. Her insurance company said no.
The lawyer asks the machine in the office: have we seen this before?
It reads every case the office has on file, right there in the building.
Forty old cases share the same situation. Each opens to the page that matters.
She reads them and decides what to do. Nothing goes out until she says so.
After eleven years, the paralegal's last day is Friday. Forty open cases sit on her desk.
A colleague opens the first one and asks: where did she leave this?
What was filed, who's waiting and what's next, pulled from her own notes, emails and letters.
It's drafted the way she wrote them. The colleague checks it and sends it herself.
find-similar-cases and summarize-case run on a local model in the building. An attorney approves anything that goes out.
Build this for your firmLocal model finds the matching cases and summarizes each file · Client documents are privileged and stay in the building · An attorney approves
Better websites, stronger visibility and connected operations need a reliable foundation. Explore the full business offer.
Try it on your business
This is just a fraction of what automation in a business can look like. Enter your business type to see how much easier the everyday work could be.
Your business model stays yours.
I help you streamline it.
Example plans, prepared with Claude ahead of time. What you type is matched to the closest example in your browser and isn't sent anywhere.
Type a sentence, or start from an example below.
Start from an example
You wrote
Example run, prepared with Claude ahead of time. Not a live call.
Preview
Skills
Example record
Runs once the map is built
The plan
I don't estimate hours here. In an audit I measure the work first, and I agree the measure with you in writing before I build.
Prepared with Claude ahead of time. This prepared example is not a recommendation for your business. A person approves every step marked below.
I build each automation from small skills. A skill does one job, is tested on its own against examples you and I agree on, and can be reused in the next automation. Local AI, on a machine in your building, does the work that touches your records. For everyday desk work with the names and personal details taken out, I can also put AI agents to work that run on the large frontier models. If one of those models is the better fit for a job, I'll tell you, and show you exactly how I'd automate it.
From example to project
I sit with your team and watch how the work actually happens. That can take days, weeks, or a month, depending on the business. I listen, follow the handoffs and learn where time is lost before recommending automation. Extended discovery is scoped in writing before it begins.
I design verification around source records, domain-specific rules and classifiers, and checks for missing or conflicting evidence. The goal is to replace long manual checks with a fast, traceable answer, not an unsupported promise.
An insurance eligibility response is not a payment guarantee. An item's value needs recent sold comparisons, condition, fees and your required margin, not just an asking price.
Proposed approach. Speed and accuracy are measured end to end for each workflow. Uncertain results are held for review, not counted as correct answers.
A named person approves anything that leaves the business.
The initial audit and written findings are free. You and I assess time, cost, frequency, customer friction and risk. I help you decide what to keep human, simplify, automate, or stop, and agree on the baseline before building.
I connect the systems you already run, build small skills that can be tested one at a time, and target less wasted time, lower cost, more capacity and better customer service. Plain rules come first; AI is used where it helps. People keep the decisions that matter.
I review the agreed measure with you after launch, comparing the live workflow with the baseline, including mistakes, review time and maintenance. The system is tested, documented and yours to operate. Then you and I choose the next bottleneck based on results, not guesses.
Opens my contact form with this plan filled in. You add your name, business and email, check it, and send it from your own email app.
Tyler B. Mishoulam · tyler@mishnetworking.com · 737-272-9916
Built with Claude.
Terms
I review your website, visibility, customer journey and daily operations with your team, then prioritize useful improvements. The audit and its written findings are free, with no commitment, and the findings are yours to keep.
Before any build, I agree with you in writing what to measure, the baseline and what counts as better. The build is quoted against the scope the audit found.
Payment follows value shown against that agreed measure. Ongoing work after handover is a retainer, scoped and priced with you directly.
Questions
Yes. Website design and improvement, search visibility, local listings, content, lead capture and follow-up are part of the offer. I connect those improvements to your booking or sales process and measure what happens after the inquiry, not just page views.
No. I do not guarantee rankings, leads or revenue. I automate repeatable work across the business, not your accountability. Your team keeps control of strategy, public content, spending and sensitive decisions. Routine messages can run under rules you approve, with consent and opt-outs respected.
No. I start by connecting what you already run. Its APIs, exports, licenses and access limits decide what is possible. Replacing a tool is a separate decision you and I make together.
Yes. The audit and its written findings carry no fee and no commitment. A build is scoped afterward, with the measure agreed in writing and payment on demonstrated value.
No. Plain rules and ordinary integrations are often enough. If the work is rare, keeps changing, or takes more effort to check and maintain than to do by hand, I will tell you it isn't worth automating.
I agree with you where data may travel before connecting anything. Local AI, on a machine in your building, does the work that touches your records. For everyday desk work with the names and personal details taken out, I can also put AI agents to work that run on the large frontier models. If one of those models is the better fit for a job, I'll tell you, and show you exactly how I'd automate it. Access is limited to the task, and a named person approves anything that reaches a customer or moves money. Local hosting removes some dependencies, not security risk or upkeep.
Before building, I agree with you in writing on today's baseline, the measure, the records it comes from, the review period and the exceptions. Then I compare the live workflow with you against that baseline, counting corrections and review time. Depending on the scope, I also track qualified inquiries, bookings, repeat business and margin after marketing, software and support costs.
Owners who need a better website, stronger visibility, more consistent follow-up or less repetitive work. The best fit can involve the people who own the work and wants to test before expanding. If you need a fully autonomous business, guaranteed savings or a replacement for professional judgment, this is not that.
Yes. I'm based in Austin, Texas, and work on site nationwide. Location, schedule, travel and scope are agreed before work starts.
It depends on the workflow, system access, integration limits and the testing needed. I set out a sequence and a schedule after the audit. There is no one-size-fits-all promise.
I remain your direct contact, day or night, for every system I build. You get the agreed operating notes and training. Support scope, response targets, third-party costs and any ongoing retainer are agreed in writing.
The workflows, integrations and operating playbook are already established. A new location can build on that foundation instead of starting over.
More locations bring more operating data. With your permission, those lessons improve the shared playbook while records and access stay location-scoped. Each rollout is configured and tested before go-live.
Where local files and control matter, I recommend in-house network-attached storage (NAS) and encrypted off-site backups. The plan includes versioned backups on dedicated on-site storage, a separately protected off-site copy, and scheduled restore tests. Extra drives or RAID alone are not a backup.
Backup locations, retention, access and recovery targets are agreed before setup. Some businesses need a different arrangement.
Local AI runs approved workloads that touch your records on hardware you control. I choose that hardware for your data requirements, speed and measured usage, not simply to add hardware.
For hosted systems, I recommend providers with a current SOC 2 Type II report and check the report scope against the services you use. That does not make MISH or your application automatically certified.
The implementation plan covers encryption, multifactor authentication, least-privilege access, monitoring and security updates. Controls and ownership are documented and verified for your setup.
Intelligence
I build websites and booking flows that connect to your CRM, then automate lead intake, follow-up and daily operations on the tools your team already uses. Local AI, on a machine in your building, is my default for work that touches your records. Someone on your team approves what matters.
One request, end to end
This is the first project I'd suggest, shown for a small refrigeration repair shop. A customer email arrives, the local model reads it, and the job runs all the way to the part, the delivery and the invoice. The local model reads and drafts. Plain rules route and match. A named person approves every spend and every new message to the customer.
Example run, prepared with Claude.Not a live call. The shop, people, messages, scores and amounts are samples.
A mail rule sends customer mail to the skill. Supplier and payment notices are filed without a model.
Walk-in cooler running warm
Hi, the walk-in cooler at our Sixth Street kitchen read 46°F this morning and it's still climbing. Can someone come out before Friday lunch? I'm on site until 3 today, or call me on 512-555-0187.
Sam Ortiz, kitchen manager
It pulls out each field and says how sure it is of each one.
| Field | Value | Confidence |
|---|---|---|
| Customer | Sam Ortiz, kitchen manager | 0.98 |
| Site | Sixth Street kitchen | 0.96 |
| Problem | Walk-in cooler at 46°F | 0.95 |
| Urgency | High: stock at risk, still warming | 0.88 |
| Needed by | Friday, 11:00 am Check | 0.72 |
| Callback | 512-555-0187, on site until 3 pm | 0.99 |
Rules, no modelAnything under 0.80 is marked for a person to check. Here that's the time: "lunch" was read as 11:00 am.
Sample scores, not measured accuracy.
Written into the job board the office already uses, through its own connection. Nothing is assigned until a named person approves.
Draft · waiting for Rosa
Waiting on Rosa
The on-call tech goes out, finds the fault and writes it on the job. A person diagnoses the cooler; the local model does not.
From the tech's note and the supplier's price list, the local model drafts the parts order and a deposit request for Sam. Both wait as drafts.
Sample amounts, not a price list.
Nothing is ordered and nothing is charged until Dana, the owner, approves both drafts.
Waiting on Dana
Plain rules do the matching. The order Dana approved is placed the moment Sam's deposit clears.
A new shipping notice is matched to the job by rule, and Sam gets the delivery day without anyone retyping it.
To Sam, Wed 8:03 amYour part arrives Thursday morning. Your tech is booked for Thursday at 2 pm.
A fixed notice Rosa approved once. A rule fills in the day and the time, and it goes out on its own.
The part arrives, the tech fits it, and the job board shows the cooler back in range.
The local model drafts the invoice from the job, with the deposit already credited. It goes to Sam only after Dana approves it.
| Diagnostic visit | $129.00 |
|---|---|
| Evaporator fan motor | $186.40 |
| Install labor, 1.5 h | $165.00 |
| Deposit paid | less $186.40 |
| Balance due | $294.00 |
Sample amounts, not a price list.
Waiting on Dana
Before any of this reads real mail
read-intakeBefore a skill reads real mail, you and I agree in writing what it should do with a set of your past messages. Every version runs against the whole set, and a plain comparison marks each answer. Three of the cases:
Clear request
Ice machine leaking at the Lamar store. Can you come Tuesday morning?
Not a request
Thanks for Tuesday. The invoice is paid.
Unclear date
Can someone come Friday? Sent on a Friday at 4:50 pm.
On these three, version 1 passes two, so it stays off real mail. Version 2 passes all three, so it's the version shown reading Sam's email in step 02.
Sample checks, not a record of a deployed system.
Where it runs.Every model step in this example runs on a local model on hardware in the shop's building, and the same agreed examples check it there.
I find the repeated handoff: intake waiting for an owner, follow-ups slipping, reports rebuilt by hand. I agree the measure with you before I build, and every workflow gets validation and an exception path.
I connect approved APIs, exports and internal tools, and build on what you already run instead of replacing it. With you, I write down the source of truth, the data mapping, and what should happen when an upstream system changes.
I build each automation from small skills. A skill does one job, is tested on its own against examples you and I agree on, and can be reused in the next automation. Local AI runs in your building for work that touches your records, and I weigh privacy, performance and budget for each task. Hardware, expected token usage and total cost are assessed together, and every candidate is tested against the same agreed examples.
Each of these is shown in full on the home page, with its own site and a worked example of its automation.
Sample business. Designed by Tyler, built with Claude.Each business listed here.
A resident's message becomes a work order. The local model sorts it with read-intake and drafts the reply and the callout; the manager approves the reply, the callout and the spend.
Rules order the materials and check the weather. read-contract and compare-suppliers run on a local model in the office, and a person approves anything that goes out.
Homes like yours are found and each protest is drafted by a local model. The owner approves every protest.
The local model writes each listing and drafts replies in your voice. A plain rule delists everywhere after a sale, and you approve each listing and every reply.
The local model finds comparable sales with find-comps, plain math sets the range, and the clerk makes the offer.
Labs are read by a local model in the clinic's building, because the clinic keeps health records on site. The physician signs every plan.
Old cases are searched and open files summarized by a local model in the firm's building, because the firm keeps privileged client files on site. An attorney approves anything that goes out.
Deliverables are finalized in the written scope.
One request-to-task workflow with a clear owner, like the run above. Before I build, you and I agree in writing how to measure it: the time from receipt to assignment, and the share of tasks someone has to correct by hand. I measure both before and after the change, and a person still reviews the exceptions.
Before anything runs, you name who approves external messages, payments and other changes that matter. I set the permissions, an exception queue and a way to stop the workflow. Every model's output is checked against examples you and I agree on, because fluency is not proof.
Free audit, and the written findings are yours to keep. You and I agree the measure in writing before work begins, and payment follows demonstrated value.
Tyler B. Mishoulam · tyler@mishnetworking.com · 737-272-9916
Infrastructure
Networks, servers, backup systems, and local compute designed around the work they need to support. I start from what you already run and change only what the work needs.
Static assembly view
Scope
Review the existing topology, device roles, remote access, and capacity. Scope configuration and upgrades to the operation rather than a generic equipment list.
Define what needs protection, restore priorities, and backup responsibilities. Include an agreed restore test and document the recovery procedure.
Compare local model options against workload, data handling, latency, and usage. Include hardware, power, cooling, subscriptions, and maintenance in the cost, not just the purchase price.
On-site compute
Local AI runs in your building for the work that touches your records. I weigh each workflow's privacy, performance, hardware, expected token usage and total cost against your budget. A person approves anything that reaches a customer or moves money. A fixed message you have approved once, like a booking confirmation, can then go out on its own.
Each local model runs on a box sized to the work: on your network, backed up and patched, with a named person looking after it.
Example run, prepared with Claude.Not a live call. A heating and air shop whose owner asked at the audit to keep customer details in the building. The shop, the call and the times are samples.
On-site model · this office
Prepared examplenames and numbersstay in the building
A missed call after hours leaves a voicemail on the office phone system.
Stays in the building
Transcribes the voicemail, then read-intake pulls out the name, address, callback number, the problem, and the system's size and age.
Stays in the building
Files the customer record on the office server.
Stays in the building
draft-reply drafts the callback text from the job facts: no cooling, a 3-ton system, 14 years old.
Stays in the building
Adds the customer's name and tomorrow's first open slot to the draft, inside the building.
Stays in the building
The text waits for you before it reaches the customer.
Waiting on you
Deliverables are finalized in the written scope.
Unreliable connectivity, untested backups, or a workload that has outgrown its machine. In discovery I connect the symptoms to a scoped fix before any equipment is specified.
Redundancy, backups, and local hosting each have failure modes. The engagement defines which checks will be performed and who owns monitoring, updates, and recovery after handoff.
That includes an on-site model: who updates it, who backs it up, and what the office does while it's down.
Free audit, and the written findings are yours to keep. I agree the measure with you in writing before work begins, and payment follows demonstrated value.
Tyler B. Mishoulam · tyler@mishnetworking.com · 737-272-9916
Advisory
A practical plan for your website, search visibility, lead generation, customer retention and operations before you commit to a tool, vendor or campaign. I agree the baseline, costs and measures of growth with you. You work directly with me, and the recommendation comes to you in writing.
Decision map
Before you commit to a tool, a vendor or a build, I map the choice with you. For each option I write down what it has to satisfy, what waiting costs, what could go wrong and who decides. Change a fact and the map checks every option again.
Example decision, prepared with Claude.Not a live call. The question is illustrative: made up, not a client, and not advice for your business. Plain rules on this page pick the recommendation, and nothing you click is sent anywhere.
The question
Should the office buy scheduling software, build on the tools it already runs, or leave scheduling alone for now?
Example: a service business books jobs by phone into a shared calendar, then types each job again into its invoicing software.
The facts
Each fact is a switch. Turn one off and the options are checked again.
Choose an option to see its criteria, what waiting costs, its risk and who decides.
Folds the three options into one written answer for these facts.
Written recommendation
Example, prepared with ClaudeThe booking steps are settled, someone re-types bookings every day, and both tools can import or export jobs. Connect the calendar to invoicing so each job is entered once.
This example runs on a local model, with the owner's permission to read the call notes. For your business, the choice of model also depends on performance and total cost. Records that touch your work stay in your building.
Agreed in writing before I build: the share of jobs typed twice, and the time from booking to invoice. If they don't move, the build stops there.
The booking steps are settled and someone re-types bookings every day, but the calendar and the invoicing software can't pass jobs to each other. Building on them would mean building around them.
The booking steps change from job to job, so anything bought or built now would lock in steps that are still moving.
Bookings are re-typed now and then, not every day, so the upkeep of a build or a new license would likely outweigh the time it saves.
You approve A recommendation, not a decision. Nothing is bought or built until you sign off.
In an engagement, an AI assistant helps me read the vendor documents and draft the comparison. I check every finding with you, on site, and the decision stays with you.
I work alongside the people doing the work. With them, I trace bottlenecks, duplicate entry, failure points, and the constraints a proposed solution must respect.
I compare integration options and total operating costs, including licensing, maintenance, and dependencies. The time horizon and usage assumptions are written down, along with what still needs testing.
When AI is part of the answer, it runs on a local model in your building. I weigh privacy, performance, hardware, expected token usage and total cost for your workflows and budget. I name who approves customer messages, payments and important changes before the workflow runs.
Findings become a build sequence with owners, access requirements, acceptance criteria, and checkpoints for continuing or stopping.
Deliverables are finalized in the written scope.
You may be considering automation, a private AI system, or an infrastructure change. A clear account of the current workflow is more useful than a predetermined technology choice.
Low volume, an unstable process, or review and upkeep that cost more than the repeated work can make automation a poor investment. A clearer process or an existing software feature may be enough.
Free audit, and the written findings are yours to keep. I agree the measure with you in writing before work begins, and payment follows demonstrated value.
About
I'm Tyler B. Mishoulam, a forward-deployed engineer and the principal of MISH Networking. I improve websites and web presence, connect customer acquisition to daily operations, and automate repetitive work so a business can grow. I come to you and build on top of the systems you already run instead of replacing them.
Austin, TX · On-site nationwide
You work directly with me, from the audit to the handoff.
I work at the intersection of business operations, software and the infrastructure underneath them. That covers workflow automation and integrations, AI skills that run on local models in your building, networks, servers and backups, and the technical decisions that come before any of it.
Forward-deployed means I come to you. I'm based in Austin, Texas, and I work on site anywhere in the country, sitting with the people who do the work, because that's where the real workflow is.
I work with AI assistants to help map workflows and write and test skills. That does not decide what runs in your business. Records that touch your work stay on a local model in your building. I weigh privacy, performance, hardware, expected token usage and total cost for each workflow against your budget. I check every finding with you before anything is built.
This site is built with Claude. How I build with Claude
Before I did this for anyone else, I did it for myself. Automating my own businesses taught me to start with the operation, not the technology: understand how the work moves, find the constraint, and build something the people doing the work can actually use.
I build on top of what's already there. Your people keep the tools they know, and replacing anything is a separate decision you and I make together.
That's still the point of the work: to free up your people's time so the business can grow and stay ready for what's next.
How I work
Five steps, from the first visit to the handover. Clear the slowest point in your workflow and the constraint moves somewhere else. That's why it's a loop.
Example workflow. Not a measured result. A real map comes out of an on-site audit of your business.
First pass · the work waits longest at estimates.Free audit
I come to you and embed with the people who do the work, where they do it. The audit is free, and the written findings are yours to keep.
Free audit
I map how the work really moves and find where it waits. I agree with you which tools and records may be read, then compare local model options for the work. I check every finding with you, and you and I agree in writing what to measure before anything is built.
Build & Deploy
I connect the systems you already run, build small skills that can be tested one at a time, and keep a person on every step that matters.
Build & Deploy
Each skill is tested on its own against examples you and I agree on, and the whole workflow against the measure written down before the build. Fluency is not proof.
Build & Deploy
Tested, documented and yours to operate. You and I review the agreed measure after launch, then decide together whether to take on the next bottleneck.
Then the loop starts again on the next bottleneck, if you want it to.
Each workflow gets its own model and hosting decision.
Free audit. Scope and measurement agreed before work begins.
Discovery, written findings, and a prioritized build sequence. No fee or commitment; the findings are yours to keep.
Quoted against the scope found in the audit. Measurement is agreed in writing before work begins; payment follows demonstrated value.
An ongoing retainer for new integrations, monitoring, and improvements as your systems grow. Scope and pricing are discussed directly.
You don't need a technical specification to start.
Tyler B. Mishoulam · tyler@mishnetworking.com · 737-272-9916
Learn
Start with the work, then choose the technology. These considerations help make the first conversation concrete.
Answer what you can. The first conversation fills in the rest.
Name the trigger, the tools involved, the people who make decisions, and what a completed task looks like. Include the exceptions that make an apparently simple workflow difficult.
Choose a measure tied to the bottleneck: receipt-to-assignment time, on-time follow-ups, or report preparation time. Before building, agree on a baseline period, source records, exceptions, and success criteria. Compare live results, including corrections and human review time.
Name the person who must approve actions that commit money, communicate externally, alter important records, or change access. Define what stops for review and how to pause or override the workflow.
Further reading: NIST AI Risk Management Framework on managing risk through design, use, and evaluation.
It may mean local inference, a private deployment, or restricted access to selected sources. Ask where data travels, who can access it, and what is logged. Neither local ownership nor cloud hosting is a security guarantee.
Further reading: NIST Zero Trust Architecture explains why location and ownership alone do not establish trust.
Start with a stable, repeated job and an accountable owner. If volume is low, source data is unreliable, or review and upkeep outweigh the work saved, fix the process first or keep it manual.
Local AI, on a machine in your building, does the work that touches your records. For everyday desk work with the names and personal details taken out, I can also put AI agents to work that run on the large frontier models. If one of those models is the better fit for a job, I'll tell you, and show you exactly how I'd automate it. Compare the options over the same period: equipment and replacement, power, cooling, service fees, backups, and staff time.
Name who patches, monitors, backs up, and restores each component, whether on site or with a third-party service. Make hardware, data, access, documentation, and third-party licenses explicit. Include instructions for operating, stopping, and recovering the agreed workflow.
Describe the handoff you’d like to improve.
Contact
Share your website address, the customers you want to reach and your growth goal. Tell me where leads or time are being lost. The free assessment can cover one issue or a plan connecting your website, visibility and operations.
This creates an email draft for you to review and send. Nothing is submitted through this website.
Open it in your email app, then send it yourself. If no email app opens, copy the draft and email tyler@mishnetworking.com.
Preparing or opening a draft does not confirm delivery.
About this site
This site is built with Claude. It's a static site: plain HTML, CSS and JavaScript files, with no framework and no npm packages.
Each claim on this page names the file in the project that backs it. Ask and I will show you any of them.
I work with Claude in Claude Code. I use it to help sketch designs, write code and tests, draft copy and run browser reviews. I set the brief and the rules and make the key calls. They're written down in the project, with the date, and so is any call I hand to Claude.
Some of the sample businesses run on small, tested functions: Lucky 7's price range and Marketplays' listing states. Others replay steps prepared with Claude ahead of time. None of them connects to a real account or calls a model live.
A review of the site as it stood lists 40 weaknesses and 26 opportunities, measured at desktop and phone sizes, by keyboard and with reduced motion. An independent review later read the new pages and the copy. Which of its findings to fix is one of the calls I handed to Claude.
A deploy uploads only the built site and its hosting settings, and mishnetworking.com changes only when I sign off. The scroll engine is Scrollcraft by Nate Herk, under the MIT license. Each sample's look started from a design reference I approved, and no logos, copy or footage from those references ship here.
Build log
From the project's review filesEach entry names the folder in docs/review/ it comes from, and every number in it is counted from the files there. All of it is dated October 6, 2026. Earlier versions of the site are not in this log.
Exploration
The site before this round, checked in Chrome from five angles: desktop, phone, speed, accessibility and robustness, and the inner pages. Single findings were then measured again, one at a time.
Source docs/review/2026-10-06-baseline/issues.json
References
Storytelling sites, agent interfaces and consultants' sites, read for technique only. No logos, copy or footage from them ship here.
Source docs/review/concepts/REFERENCES.md
Concept sketches
The inference engine, the business mind and you're in the chair. Each is one self-contained page with notes and screenshots.
Source docs/review/concepts/
Critique
The reviewer kept its own frames. The inference engine scored 7 for craft, 7 for clarity and 8 for thesis, with a must-fix list. The other two record a revision made after it.
Source docs/review/concepts/a-inference-engine/CRITIQUE.json
Decision
The business mind opens the home page, the sample lineup follows, the day in the chair is the climax and the inference engine closes.
Source docs/review/DECISIONS.md
Round 1
A browser test suite with set limits, a build that ships only the files on its list, contrast and focus fixes with before and after shots, the film and share image, and a Pause motion control. Then the round was reviewed.
Source docs/review/2026-10-06-round1/
Copy and identity
Every visible string tagged new, changed, kept or dropped, prepared with Claude for my approval. One type pair, one signal color and a paper layout for pages like this one, with every allowed color pair measured.
Source docs/review/copy-deck.md docs/review/identity-spec.md
Rounds 2 and 3
The business mind, the day in the chair and the inference engine, each rebuilt as a piece of the page with its own test page.
Source docs/review/ports/
Round 4
The pages behind the home page move to the paper layout, each with a test page and screenshots. This page was drafted in this round.
Source docs/review/ports/inner/
Review
A separate reviewer read the new home and inner pages and the copy deck, and filed code findings, copy notes and a search audit. Its notes on this page's own claims are applied here.
Source docs/review/hermes/
Client systems are chosen separately from the tools used to build this site: local AI in the client's building for records that touch their work. A person approves what matters.
docs/review/DECISIONS.md docs/review/copy-deck.mddocs/review/DECISIONS.md docs/inspiration-map.json docs/review/concepts/site.mjs tests/docs/review/2026-10-06-baseline/ docs/review/hermes/docs/review/DECISIONS.md STATE.mdStatic files, no framework, no npm packages.
index.html, holds every page. The end of the web address, after the #, picks the one that shows.package.json lists five scripts and no packages.assets/fonts/.vercel.json.tools/build.mjs copies exactly the files listed in site-manifest.json into dist/, and nothing else. It stops before touching dist/ when a listed file is missing or empty, or when a page on the site asks for a file the list leaves out. It also refuses to ship notes, tools, tests, backups and .env files, even if someone lists them.
.vercelignore limits the upload to dist/ and vercel.json. tests/deploy.test.mjs builds into a temporary folder and simulates that upload.
Two suites: one in Node, one in a real browser. Neither uses npm packages. The first needs only Node; the second also needs Chrome and the site running on the same computer.
Some samples decide things with small functions in site.mjs. They work on example data and connect to nothing. Each one has a test, quoted here by name.
crossList()Marketplays. Marks an item sold where it sold and delisted on every other marketplace.Test: "one sale delists the item everywhere else"
money()Lucky 7 Pawn. Formats the dollar amounts on the loan ledger.Test: "money formats ledger amounts with a sign and cents"
plumbCap()Plumbline Construction. Maps the scroll position to one caption at a time, its scene, and how far the row of dots has lit, and back again. Its module is plumbline-story.mjs.Test: "the scroll-to-caption mapping is monotonic, reversible, reaches every caption, shows one at a time and ends on the last beat"
valuation()Lucky 7 Pawn. A price range from example sold prices for similar items, then buy and loan offers below the middle of it.Test: "pawn offers stay below the comparable median"
All ten sample businesses appear as full-depth prepared examples. Their rules stay in site.mjs with their tests. The utility adaptSession() is also tested. Current exhibits include: dispatchPlan() for Solstice's battery, and nextDrop() and countdown() for Scrapyard's drop timer.
The rest of site.mjs runs the site itself. resolveRoute() handles the page links. buildInquiry() and contactContext() write the contact email draft and name the sample a visitor came from, carryRegarding() keeps a Regarding line the visitor rewrote, and planHandoff() and planMessage() carry a plan from the plan builder into that draft without cutting what the visitor wrote. createFilmController() runs the CPU film, and initialMotion() and createMotionClock() run Pause motion.
tests/showcase.test.mjs32The local preview and the deploy both serve exactly site-manifest.json (tests/deploy.test.tests/opening-motion.test.mjs24Checks particle geometry, scroll scrubbing, reverse motion, pause/resume, resizing and static fallbacktests/engine.test.mjs20Inference engine (the home page close): matcher, plan schema, prepared data, run scripts, tests/owner-day.test.mjs19Checks in owner-day.test.mjstests/inner-infrastructure.test.mjs17The Infrastructure page: its markup, copy rules and moduletests/inner-intelligence.test.mjs15The Intelligence page: its markup, copy rules and moduletests/site.test.mjs15Codex finding 3 (docs/review/codex/CODEX-REVIEW.md): following a sample's contact link reptests/motion.test.mjs12Astra review finding 2 (docs/review/codex/ASTRA-ULTRA-REVIEW.md): Scrollcraft's pointer lotests/chrome.test.mjs13The MISH chrome around every page: the sticky channel bar, its guide, the footer and the rtests/day-close.test.mjs11Part 3 of the integration round: the day walkthrough (concept C), the engine close (conceptests/day-window.test.mjs7The box as one window: every word in order, one slide at a time, a short scroll and the plain stack with motion off.tests/deploy.test.mjs11"Part of the site" is what a visitor or a crawler can reach: everything index.html loads, tests/live-metals.test.mjs11Fixed owner-supplied response is a unit fixture, never a fallback quotetests/seo.test.mjs11Search and answer-engine files (docs/review/hermes/seo, applied per docs/review/DECISIONS.tests/inner-advisory.test.mjs10The Advisory page: its markup, copy rules and moduletests/inner-how-i-build.test.mjs9The How I build page: its markup, copy rules and moduletests/lucky-neon.test.mjs9Checks in lucky-neon.test.mjstests/fairweather-story.test.mjs6The Fairweather section: its wording in order, what it does not promise, one caption at a time, the camera, a wall of exactly a thousand folders with forty lit, and that the old contract files are gone.tests/porchlight-story.test.mjs6The Porchlight section: its place on the home page, its wording in order, no promised result, one caption at a time, the camera pulling back from the notice to the town, and the street and town data.tests/plumbline-story.test.mjs10The Plumbline section: its wording in order, the scroll-to-caption and camera mapping, the plain list for reduced motion, the height budget, and the copy rules.tests/check-site.test.mjs8Verdict logic of the browser suite (tools/check-site.mjs, issue 31), without launching Chrtests/inner-pages.test.mjs7The Pages page: its markup, copy rules and moduletests/marketplays-story.test.mjs7The prepared Marketplays story separates possible channels from a category-matched exampletests/mobile.test.mjs7Phone layout guards (round 1, mobile area)tests/business-growth.test.mjs6Checks in business-growth.test.mjstests/seams.test.mjs6Every top-level declaration block for exactly this selector, joined (a selector may be stytests/cinematic-contract.test.mjs5Checks in cinematic-contract.test.mjstests/head-a11y.test.mjs5Checks in head-a11y.test.mjstests/mish-backgrounds.test.mjs5Checks MISH-only backgrounds, visibility, pause and reduced motiontests/opening-provider-choice.test.mjs5Checks in opening-provider-choice.test.mjstests/pawn-showcase.test.mjs5Checks in pawn-showcase.test.mjstests/payoff.test.mjs5Section 9 replaces head/pace fast-forwarding with construction + narrative clockstests/phone-control.test.mjs5Checks lazy video playback, motion controls, poster fallback and interrupted playbacktests/ashbury-environment.test.mjs4Dependency-free DOM boundary; browser paint/geometry is checked by the parent harnesstests/broad-recovery.test.mjs4Checks in broad-recovery.test.mjstests/cinematic-motion.test.mjs4Checks in cinematic-motion.test.mjstests/hale-anatomy.test.mjs4Dependency-free DOM adapter; assertions inspect every mutation of real render outputtests/hale-biology.test.mjs4Checks in hale-biology.test.mjstests/hale-story.test.mjs4Checks in hale-story.test.mjstests/kitco-reference.test.mjs4Checks in kitco-reference.test.mjstests/lucky-tickets.test.mjs4Checks in lucky-tickets.test.mjstests/paused-arrival.test.mjs4Exercise the real play/wake and Lucky renderer, not a rewritten lifecycletests/boot-integration.test.mjs3Checks in boot-integration.test.mjstests/business-branches.test.mjs3Checks in business-branches.test.mjstests/color-motion.test.mjs3Checks in color-motion.test.mjstests/mind-framing.test.mjs3A complete composition, rather than a viewport-wide graphic hidden by the copy masktests/spec-contract.test.mjs3Build spec numeric typing and pure scroll ownershiptests/business-offer.test.mjs2Checks in business-offer.test.mjstests/pawn-workflows.test.mjs2Checks in pawn-workflows.test.mjstests/phone-founder-copy.test.mjs2Checks in phone-founder-copy.test.mjstests/ashbury-ground-up.test.mjs1Checks in ashbury-ground-up.test.mjstests/engine-journey.test.mjs1A single patient record must travel through the manual story before its assisted counterpatests/pawn-network.test.mjs1Checks in pawn-network.test.mjstests/tradingview-market.test.mjs1Checks in tradingview-market.test.mjstests/videos.test.mjs9The Videos page: its route, nav and footer, the tiles from videos.json, the self-hosted thumbnails and the sync script's pure partstests/build-loop.test.mjs15The Office to model section: its place on the home page, its wording, the plain stack it falls back to, the scroll timeline, the compound beat's order, camera, brain and colors, and its film frames.tests/local.test.mjs11Checks the Why local section: its place on the home page, the agreed wording, the text alternative, the motion rules, how many grains it draws and the files it ships.node --test tests/*.test.mjs443 tests in 57 filesRun
tools/check-site.mjs opens the site in Chrome with no window on screen, at laptop size (1440 by 900) and at phone size (390px wide). Every limit is set in one place in that file, and the run fails when any is missed. Its pass and fail rules have their own tests in tests/check-site.test.mjs.
site-manifest.json.assets/scrollcraft/, and its license ships with the site.assets/fonts/.docs/skill-inventory.md.Free audit, and the written findings are yours to keep. I agree the measure with you in writing before work begins, and payment follows demonstrated value.
Tyler B. Mishoulam · tyler@mishnetworking.com · 737-272-9916
Videos
How I build local AI for a business, one episode at a time.
Newest first. Each one opens on YouTube in a new tab.
Every episode is on the MISH Networking channel. Open the channel on YouTube
