LexAmp

LexAmp was an attempt to build a marketplace connecting people who need routine legal help with lawyers who could provide it, using AI to handle the parts that do not need a lawyer. It came out of the UofT Rotman Commerce Boardroom competition in 2024 and was developed further for the Diamond Challenge and FBLC.

It is no longer active. I am leaving the page up because the reasoning behind it is still worth reading, and because retiring a project quietly is worse than saying why.

The problem it was aimed at

Most people who run into a legal issue never talk to a lawyer, and cost is the usual reason. On the other side, small firms spend a large fraction of their time on administrative work that does not require legal training. The bet was that a lot of what sits between those two facts is templatable: intake, document assembly, first-pass triage of whether a matter actually needs a professional.

What was built

AI legal chatbot. Handled initial consultations, assessed whether a matter was simple enough to answer directly, and escalated to a human lawyer when it was not. The escalation logic was the interesting part -- knowing when to stop is harder than answering.

LexCraft. A contract generation system that selected and populated templates from a plain-language description of what the user needed, with a paid tier that routed the output to a partner firm for review.

Stack: Next.js with React and TypeScript, Prisma over PostgreSQL, NextAuth for identity, and a template-based generation layer with NLP for intake.

Why it stopped

It never had real users. The competition version was a working prototype and a business model, and business models win competitions, but the gap between "a panel of judges finds this plausible" and "a law firm changes how it operates" turned out to be the whole problem. Legal services are a regulated, trust-dependent, relationship-driven market, and none of those three things are helped by a marketplace with no supply side.

I also came to think the product was solving the wrong half. The bottleneck for a small firm is not document generation; it is client acquisition and the liability attached to anything the firm puts its name on. An AI that drafts faster does not reduce the second one at all.

What it was worth

The technical work carried forward -- the retrieval and template-assembly patterns showed up again in later projects. The more useful lesson was about the distance between a validated pitch and a validated product. A business plan that places in three competitions has been validated against judges, not customers, and I spent longer than I should have treating the first as evidence of the second.