AI matcher (/find)
Natural-language search for AI builders. Claude Haiku reads your query against our case graph and returns five builders with a reason for each.
What it does
You write a paragraph in plain English. The matcher returns up to 5 builders whose published cases match — with a specific one-sentence reason per match.
Under the hood
- Model: Anthropic Claude Haiku 4.5
- Inputs: your query + the builder catalog (handles, stacks, case titles + summaries)
- Output: top-5 builder handles + reasoning
- Latency: typically < 5 seconds
Defense in depth (8 layers)
We defend the endpoint heavily so it stays free + abuse-proof:
- Origin check — same-site only
- Input sanity — length, encoding, content
- Rate limit — per-IP, daily-salted hash (no plaintext IPs stored)
- Turnstile — invisible CAPTCHA, warmed on focus
- Query cache — identical queries reuse cached LLM result (~24h)
- Daily budget cap — $5/day at Anthropic, hard ceiling
- Relevance gate — non-builder queries (gardening, recipes) get a polite refusal
- Content guard — obvious-harm queries refused before reaching the model
Why this is our differentiator
Traditional career platforms force you into keyword-style search or recruiter-mediated discovery. Neither works for AI-builder work where the stack matters more than the title. Plain-language → real people is the right shape.
Try it
Go to /find and type a real problem.