What I've built

Three systems in detail — including the numbers that didn't improve.

The clinic that was losing a dozen calls a week to a phone nobody could answer

Dental clinic, Bangalore · AI voice receptionist · live since 30 July 2026

Before. Two dentists, one receptionist on the desk. Calls came in while she was with a patient, after closing, and on Sundays when the clinic was shut. Ten to fifteen a week went unanswered. Every one of them was someone trying to book.

What I built. A voice agent on the clinic's normal phone line. It answers every call, books into the calendar, and manages follow-ups for patients on multi-visit treatment plans. When someone misses an appointment it rings them to ask why, and messages them on WhatsApp.

The part I'm proudest of. The dentist updates it by talking to it. He sends a WhatsApp voice note — “I'm out Thursday, emergency” — and the agent updates its own knowledge, reschedules the affected appointments and messages those patients itself. No dashboard, no form, no waiting for me.

After. Unanswered calls: effectively zero. Bookings up 30–40%. No-show rate: unchanged — I'd rather tell you that than pretend everything improved.

Built in three weeks. Running since July. Around two hours a week to maintain, and every account and subscription sits in the clinic's own name, not mine.

In progress

The coach who was getting more enquiries than anyone could answer

Fitness coach · 1-to-1 coaching, live and recorded courses · WhatsApp + email · in build

Before. A large Instagram following generating more enquiries than his team could reply to. One or two assistants handling onboarding, session links and invoices by hand — for live sessions with 200–300 people and recorded courses with thousands. Clients in several countries, so follow-ups were being lost across timezones. The problem was never demand. It was that demand was arriving faster than anyone could answer it.

What I'm building. One system that takes someone from first enquiry to paid: replies, onboarding, a meeting link for every session, invoices, and payment reminders — on WhatsApp and email, wherever that person actually responds.

Status. In build. The hold-up isn't the automation — it's that the information it needs lives in several people's heads, and we're organising that first. With everything in place it is about twenty days of work. That's normal, and it's worth saying out loud: most of the delay in automation projects is never the automation.

A finance product, built alone in two months, live with real users

AI SaaS · finance · subscription product · three-year contract

What I can say. It's an AI product in finance, sold on monthly subscription, used by both businesses and individuals. It's live with real paying users. I built it in about two months.

What I can't. The idea itself. That's the client's and it stays theirs.

What I'm contracted for. Three years of building, maintaining and improving it, including preparing it for enterprise customers with an API so other companies can build on top of it.

How I'm paid for it. I took a share of the revenue instead of a fee, because I thought it would work. That means I only do well if the product does — which is also why I'm still building it three years out rather than handing it over and moving on.

Why it's on this page. Everything else here is automation — finding a bottleneck in a business and replacing it. This was different: a product from nothing, security included, with no engineering degree behind me. It's also why I know what “maintainable” means, because I signed up to maintain this one for three years.

Something in your business work like this?

Tell me how your operation runs and I'll tell you honestly which parts are worth automating — and which aren't.

Book a free 30-minute call