π‘οΈ Guardrails that retry instead of just blocking ποΈ Classifiers that only call an LLM when they have to π RAG that gets measured, not just shipped
π₯ Now: hybrid search + reranking, with evaluation built in π€ Open to: applied AI roles and collabs
π 41 Total PRs Β Β β’ Β Β π’ 23 Merged PRs
