The demo takes a weekend. Then someone asks whether the assistant can show a user something they shouldn't see, and what happens the first time it takes an action — and the project stalls for three months.
I build the version that answers those questions. Seventeen years architecting and delivering software; the last few building AI systems that reach production.
An internal policy assistant with role-based access enforced at retrieval time and human approval before any write action. Sign in as two different users, ask the same question, get correctly different answers.
Full source, tests and architecture decisions on GitHub → github.com/compcode-solutions/contoso-policy-assistant
Automated web regression testing. Point it at a running app; it explores, captures a baseline, and catches UI and behaviour regressions on every release.
No sales call required to evaluate either of these.
I'm taking on three founding clients at half my list rate. In exchange I ask for a testimonial, permission to write the work up anonymised, and an introduction if you're happy with it. Both prices are shown below — after three, it's list.
Two working days inside your codebase and your workflow. You get a written report: where AI genuinely helps you, where it's hype, what it would cost, and what to do first. No pitch at the end — the report is the deliverable.
3–5 days · ₹50,000 / $1,100
Founding-client rate — list ₹1,00,000 / $2,250. First three clients only.
Your docs, policies and wikis become a chat interface your team actually uses — with role-based access so people only see what they're cleared for. Deployed to your infrastructure.
2 weeks · ₹2,10,000 / $6,750
Founding-client rate — list ₹4,25,000 / $13,500. First three clients only.
Give us a URL. We explore your app automatically, capture a baseline of how it looks and behaves, and check every release against it. You get a report showing exactly what changed — before your customers find it.
2 weeks · ₹1,75,000 / $5,250
Founding-client rate — list ₹3,50,000 / $10,500. First three clients only.
Optional: ₹21,000 / $425 per month to keep the baseline current as you ship.
An AI assistant embedded in your product, trained on your data, with the guardrails needed to put it in front of real customers.
3–4 weeks · ₹4,25,000 / $13,250
Founding-client rate — list ₹8,50,000 / $26,500. First three clients only.
Some engagements need more than one specialist for two weeks. I've spent seventeen years architecting and leading delivery teams, and I assemble one when the work calls for it.
How this works, stated plainly: I don't keep a bench. Teams are assembled per engagement from people I've worked with and would vouch for, against a signed scope. That means I can't start Monday with six people — and it also means you're not paying for anyone's idle capacity, and nobody is rotated onto your project because they happened to be free.
Most clients start with a sprint from the section above. This is the conversation that follows.
Seventeen years building software; the last several building AI systems that reach production rather than stopping at a notebook.
The pattern I keep meeting: a team knows they want AI in the product, they've seen the demos, and nobody in the building has taken one from prototype to something real users touch. The hard parts aren't the model. They're retrieval over messy real documents, permissions that match your actual org chart, guardrails on anything that writes, and knowing whether the answers are any good.
I write about this work at /blog. If you want to know how I think before you hire me, start there.
I ask about your situation. If I can help I'll tell you exactly how; if I can't, I'll tell you that too and point you somewhere better.
within 24 hours — deliverables, timeline, price, and an explicit list of what's not included.
(100% on the audit — it's small). Work begins on the agreed date.
working software throughout. No four-week silence followed by a reveal.
session, runbook, and everything in your repositories. You own all of it.
Often you should, and I'll say so on the call. Off-the-shelf helps an engineer who is already writing the code. These engagements are for when there's no coverage to begin with, or when the system needs to know who's asking before it answers.
That's what the audit is for — a few days and a small number, before anyone commits to two weeks.
Yes. Everything transfers on final payment. I retain rights only to my own general tooling that predates the engagement.
Usually. Everything can be deployed to your own infrastructure, and access control is designed in from the start rather than bolted on.
Both, honestly. I do the specialist work myself. For larger scopes I assemble a team from people I've worked with, against a signed engagement — I don't keep a bench and I won't pretend otherwise.
Yes — fractional architect, tech lead on retainer, or extended delivery. Ask on the call.
25 minutes. No deck. If it isn't a fit, I'll say so and point you somewhere useful.