qv* / project

Reflect AI Search

“A semantic search layer for a 300k-note knowledge base.”

2025 / for reflect / in-progress

180msp95

query latency

0.87

recall@10

the problem

Reflect's users had three years of notes and could only search by keyword. Long-tail queries returned nothing useful.

the solution

I built a semantic search layer on top of their existing storage. Notes are embedded on write, queries are routed through a small ranking model. Results return in under 200ms.

human contribution

I prototyped the embedding pipeline myself. AI handled the API surface and the test harness.

built with

cursorclaude-codetypescriptpostgrespgvectoropenai

Hire Maya for something like this.

open-to-work · $250/h · next.js

hire Maya →view full profile

more by Maya Okafor

acme pricing dashboard

A pricing engine the sales team actually trusts.