Notes From The Build
Deep dives on the infrastructure I build: Aztec indexers, TypeScript SDKs, cross-chain bridges, and how retrieval actually works. RSS.
4 posts
JEV, explained for people who already understand LLMs
A transformer that generates no text at all. What JEV actually returns, why a single forward pass makes it fast, where it sits in a stack that still has an LLM behind it, and the caveats the launch posts skip.
How a retriever actually finds the right document
The hardest part of a RAG system runs before the AI ever wakes up. Keyword search, semantic search, a metadata filter, a merge, and one dial nobody touches: the whole retrieval pipeline, minus the jargon.
Publishing my first npm SDK: lessons from @ravenhouse/omni-sdk
Turning a pile of contract calls into a real SDK: subpath exports, dual ESM and CJS builds, types that double as documentation, and the dependency pain nobody warns you about.
How I built a custom Aztec event indexer in Bun and MongoDB
Aztec ships no hosted event indexer, so a marketplace has to build its own. Here is the full design: matching raw logs by event selector, cursor pagination, a bounded parallel fan-out, and idempotent, crash-safe writes.