About
An AI lead qualification and ranking system that finds the best contacts at target companies against an ideal customer persona. Built as a technical challenge.
Contacts import from CSV, get qualified, then rank within their company. The ranking prompt optimises itself rather than being hand-tuned.
Key Features
- Two-phase pipeline. Qualification runs first, then ranking, so cheap filtering happens before expensive comparison.
- Automatic prompt optimisation. Beam search explores prompt variants instead of relying on hand-tuning.
- Per-company ranking. Contacts rank within their own company rather than one global list, which keeps comparisons meaningful.
- Background job abstraction. The same tasks run synchronously in development or through a queue in production.
- Model flexibility. OpenRouter handles routing, so the model can change per task without touching code.
- Safe defaults on AI failure. A model error degrades the result rather than breaking the pipeline.
Tech Stack
- TypeScript
- TanStack Start (SSR)
- React 19
- TanStack Router / Query / Table
- ORPC
- PostgreSQL
- Drizzle ORM
- OpenRouter
- Trigger.dev
- Jotai
- shadcn/ui
- Tailwind CSS 4
- Turborepo + Bun workspaces
Video
Source Code
Screenshots
