Sentinel
Context-repair layer for production RAG. Catches hallucinations via citation cross-checks and stitches in missing sources before they reach the user.
Things I've designed, shipped and maintained — across AI, data engineering, systems and open source. A few are live, a few are archived, all are real.
Context-repair layer for production RAG. Catches hallucinations via citation cross-checks and stitches in missing sources before they reach the user.
Debezium → Iceberg with schema-aware merges. Handles 60M rows/hour without blinking and never loses a row on replay.
Tiny Rust static-site generator built around Pandoc. Incremental builds, typed front-matter, ~2k LOC. Powers this very site.
Local-first notes with CRDT sync. Markdown, backlinks, full-text search, end-to-end encrypted, and no server when you don't want one.
Near-real-time feed analytics for a 200M-user consumer app. Spark Structured Streaming feeding ClickHouse for sub-second dashboards.
Personal dotfiles. Curated, minimal, boringly reliable — and somehow 200+ stars. The boring infrastructure of a productive terminal.
Every project name is a real word. More lives on github.com/eternalAbyss · linkedin.com/in/us9044
I'm open to AI / ML roles and collaborations. Grab my résumé or send a note.