AI features that do a real job inside your product, or a whole AI product built from scratch.

That can mean chatbots and WhatsApp agents that answer from your own content, agents that research and act on a schedule, AI that matches, drafts or scores, or voice interfaces. I build them into a real product with a real interface, handle the prompts, retrieval and guardrails, and test them against real examples before anyone relies on them.

A working AI feature or product, running for real users rather than as a demo

The model and approach chosen for your case: OpenAI, Claude or open models, with retrieval over your own data where it helps

The interface, backend and integrations around it

Testing against real examples before launch, so you know where it breaks

Usage and cost tracking, so the model bill doesn't surprise you

Who this is for Founders adding AI to a product they already have, or building an AI product from the ground up, who need it working for real users.

How it runs A short scoping call, then a written plan. Where the idea is unproven, the first phase is a small working version on real data, so we find out quickly whether it holds up before building the rest. A first working version usually takes about four weeks.

Recent: Parcoursup Zen, which matches French students against 23,000+ formations, drafts motivation letters and runs an orientation chatbot. Chatberry, an Arabic-first WhatsApp platform with an AI chatbot and 1M+ messages sent. techpotions.app, an AI agent that finds businesses showing buying signals on the live web.