I build AI systems that are reliable, observable, and intelligent. RAG engines, AI agents, and LLM observability platforms — designed for production, not demos.
Adaptive RAG Engine — RAG system that dynamically selects its retrieval strategy per query. Hybrid search (vector + BM25 + RRF), self-evaluation pipeline, semantic chunking, and refusal path. 88 tests.
Agent Framework + DevTools — Lightweight framework for building AI agents with full execution traces. @tool decorator, ReAct + Planning modes, SQLite trace persistence, memory system, and Trace Explorer UI. 69 tests.
AI Pipeline Observatory — LLM observability SDK. Auto-instruments OpenAI/Anthropic with 3 lines of code. Cost tracking, quality scoring, hallucination detection, and EMA-based drift alerts. 28 tests.
The three projects form an ecosystem: Observatory instruments the other two, providing a unified monitoring layer across RAG and agent pipelines.
AI/ML: Claude API, OpenAI Embeddings, Qdrant, BM25, Cohere Rerank, RAGAS evaluation
Backend: Python 3.12, FastAPI, Pydantic v2, SQLite, SSE streaming
Frontend: React 18, Tailwind CSS, Recharts, Vite
Infra: Docker Compose, Nginx
- Telegram: t.me/kazdoveth
- X: @kazdovETH