I design and build evaluation-driven search & AI systems, specializing in the intersection of Optimization, Vector Search and Data/AI Pipelines using testing, containerization, monitoring, logging, and guardrails.
Advanced Search & Re-Ranking
Elasticsearch, Solr, Vespa.
- Research for Sease - Integrated Vector Search support using Solr DSL in RRE (Rated Ranking Evaluator).
- Hybrid & Dense Embedding strategies for domain-specific retrieval (Legal/Healthcare).
- Contributions: Quepid Integration / RRE Dataset Generator.
Vector Search / RAG
FAISS, Haystack, Milvus, LangChain Integrations.
- LLM Consensus: LES: Legal Engine Search.
- RAG Systems: building agents & LLM apps with focus on reducing hallucinations with factical grounding. rag-prototype | rag-openai-chats.
LLM Infra and Tooling
Efficient guardrails with CI flows with cache, tags and labeling.
- RepoGPT: repo summarization. Structured codebase summary for human, LLMs and RAG systems.
- MCP Local Agent: local agent with custom tooling via local MCP.
- LM-Stacks: LLM resources and tooling.
Experimental (ML)
PyTorch, NetworkX, Scikit-Learn.
- Geometric-Aware Retrieval (Experimental - manifold curvature in embedding spaces for non-euclidean retrieval).
Research & Applied Science (Game Theory & Reinforcement Learning)
- Board Game Solvers: minimax/MCTS implementations (Chess Solver).
- RL: Deep CFR & NN approaches for Blackjack (Blackjack_vs_Naive).
- Tooling: custom Gymnasium (based on OpenAI original idea).
- OSINT: Sandworm-Spain Analysis.
(Always up for a chess or a poker game!)