A legal platform for my mom's firm
MCPC Advocacia, live since 2026
My mom is a lawyer. Her firm is the legal arm of Razonet, my family's accounting firm, and it handles trademark registrations and contracts for small businesses. Until this year, she did all of it by hand.
- Built by
- me, solo
- Processed
- ~30 registrations and contracts
- Behind the AI
- ~4,000 contracts
The problem
Legal work only happened when a client asked for it, and every request meant emails, documents and payments handled manually by one person. It didn't scale, and it wasn't a product anyone could buy.
What I did
I sat with my mom and mapped every step of how she works. I drafted a business model, selling to Razonet's clients and to anyone else, and got my parents to approve it. Then I cut it down to an MVP and built it alone, coordinating with her, the accounting team and the developers at Razonet.
- Client and firm dashboards, with login and permissions per client.
- Sign up with a Brazilian tax ID that fills in the company data automatically.
- Payments directly on the site.
- Security basics done properly: row level security, rate limiting, security headers and idempotent webhooks.
The AI part
My first attempt used vector search over the firm's documents, and it was bad. It found similar text but didn't actually work the way the firm works. So I rebuilt it around a curated index: the firm's rules, templates and clauses written down, plus about 4,000 past contracts organized by category. The model reads the index, picks the three to five documents it needs and drafts from them, with the newest contract winning when clauses conflict.
The difference is simple. Vector search finds something similar. An index teaches the model how the firm does things.
Stack
TanStack Start (React with server rendering), Supabase (Postgres, Auth, Storage), Cloudflare Workers, and the Claude API.