Bilal Chafi

I build AI systems that answer, decide and automate, and I keep the trail of how they got there. Every project here has a journal: what I tried, what broke, what I chose and why.

Activity

1 Oct
designReview IntelligenceCover: from "stars vs words" to the hidden return→
27 Sep
designReview IntelligenceNo prompt: facts in, typed questions out→runReview IntelligenceRun 1: right answers, wrong questions→runReview IntelligenceRun 2: over-correction, and my rule hiding the flags→runReview IntelligenceRun 3: stable, a report a founder can act on→

Projects

Review Intelligence

Star ratings say how a customer feels, not what went wrong. A sarcastic 5★ review can hide a refund. I built a pipeline that reads every review and returns typed answers a business can count.

5 changes→

AI research team in a real browser

A client needed verified contacts (site, email, phone, director, LinkedIn) for thousands of training organisations. Agents had to find them on the open web without inventing anything.

→

Lead scoring from an ICP

Score inbound leads against an ideal customer profile with evidence, not vibes: one research agent per question, a source link for every answer, and a score computed by code.

→

Contact

GitHub
Bilal Chafi

I build AI systems that answer, decide and automate, and I keep the trail of how they got there. Every project here has a journal: what I tried, what broke, what I chose and why.