Search Engine for Internal Docs
Built an internal semantic search engine over 2.4M company docs using local embeddings and a custom HNSW index. Query latency p95 under 80ms.
- Rust
- Postgres
- pgvector
- Tantivy
- Cloudflare Workers
A few sentences about who you are, what kinds of problems you take on, and what you care about. Be specific. Generic bios get skipped by humans and indexed flatly by AI.
Most of my career has been at the seam between resilient backend systems and the people who depend on them not falling over. The work I'm proudest of is invisible — infrastructure you only notice when something breaks elsewhere.
I write about what I learn at /articles and keep an honest log of the projects I started and didn't finish in the archive at the bottom of /projects. Both feel like an essential counterweight to the highlight-reel version of a portfolio.
If you want the snapshot of what I'm focused on this season — what I'm building, reading, learning — that's at /now. Updated whenever it stops being true.
Lead engineer on the platform team. Designed the multi-region failover system that now handles 12M requests/day.
Joined as a senior IC, promoted to tech lead after the platform reorg. Shipped the v2 event-replay system.
Backend engineer on the data ingest team. Owned the schema-evolution layer and the customer-facing event-replay tooling.
Built an internal semantic search engine over 2.4M company docs using local embeddings and a custom HNSW index. Query latency p95 under 80ms.
A 12M-req/day event ingest and replay system built on Kafka and Postgres, with per-tenant schema evolution and deterministic replay.
A terminal tool that streams pg_stat_statements into a live table and lets you pin queries to a hotlist. Open source, ~600 stars.