Start wherever you are.
Six rungs, smallest first. Each one stands on its own. Start where you like and stop when you want. The point is to learn whether the work is worth doing before you spend real money.
Each rung stands on its own.
Start where you like and stop when you want. Two rungs carry no figure, because I price those after discovery rather than before it.
Find out what is worth doing
Answer about fifteen questions about how your business runs today. You get back a ranked shortlist. It shows where AI pays off and where it flops. It shows what each one costs you to build. One answer on that list is to fix a process and buy nothing.
Roadmap Lite
We take your scan results into one working session. You leave with a written map of three opportunities. It names what to do first. It names what that one is worth. It names what your data has to look like. It names what the build takes. The document works whether or not you hire me.
Second Opinion
Your AI project stalled. I take it apart and tell you why. I report what got built. I report why it falls short. I report whether it is worth saving. I run an evaluation harness against it so the answer carries numbers. Sometimes the answer is that your vendor did fine. The data was the problem. You get the truth either way.
The AI Opportunity Map
Paid discovery. I interview your team and audit your workflow and your data. I clock how long the work takes today so later savings measure against something real. You get a ranked roadmap. You also get a working prototype of the top opportunity. The recommendation shows rather than argues. I price it once the scope is clear, because quoting before discovery is guessing. If the honest answer is that you should skip AI, the roadmap says so.
The Build
Fixed price, milestone by milestone. You approve written pass/fail criteria before each one starts. The invoice arrives after the work passes them. A staging link stays live the whole way through. The contract caps your time at about an hour a week. I price it against the scope the Map produces. The Pricing page explains why no figure sits here.
The Monthly Proof
Every system I build ships with a meter. The first ninety days are included. After that the meter keeps running. Each month you get a report of what the system handled and what it cost and saved. I watch the models and APIs it leans on, so you schedule a deprecation instead of discovering one. Your evaluation suite runs on repeat, so quality drift pages me first. Monitoring runs on systems I built or took over.
What this stack is wrong for.
Three things I route elsewhere or flag during discovery. A stack pitch nobody qualifies is just a sales pitch.
Long-running compute
Cloudflare Workers cap CPU time. So work that runs for minutes belongs outside a request. I handle it with queues or Durable Objects or a Python sidecar called from the edge. I flag it during discovery while it is still a design choice. Finding it mid-build turns it into a surprise.
Heavy media processing
Video transcoding and large image pipelines belong on services built for them. I route that work out rather than force it onto this stack. The integration is mine. The processing is theirs.
Problems this stack is wrong for
Some problems fit other tools better. The honest move is to say so rather than reshape your problem until it fits what I know. If yours is one you hear it in the first conversation. That beats hearing it in month three.
Open for projects
Find the work worth doing.
Tell me what runs slow or costs too much or keeps breaking. If AI is the wrong answer I say so.
