I build and design evaluation-driven search & AI systems. Enjoying the intersection of Optimization, Vector Search and Data/AI Pipelines. Focused on:
Search & Re-Ranking
Elasticsearch, Solr, Vespa.
- LLM Consensus. LES: Legal Engine Search; Web PDF (Open), UPM (Auth Req.).
- Research for Vector Search support in RRE (Rated Ranking Evaluator) using Solr DSL.
- OSS Contributions: Vespa, Quepid, Pytests, Git CI, Testing and Integrations with Docker (among others).
Retrieval / RAG
FAISS, Haystack, LangChain.
- Modular RAG (Custom Ports & Adapters Framework).
- Using Milvus to process OpenAI chats and local RAG.
- Hybrid vs. Dense; Search vs Retrieval. Benchmarking (BEIR, MTEB, RAGAS).
- LoRA for Contrastive/Triplet-Loss on Legal/Healthcare domains.
Tooling and Agents
LLMs, local MCP.
- 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)
GraphRicciCurvature, NetworkX, Scikit-Learn.
- Geometric-Aware Retrieval (Experimental Sandbox - working on manifold curvature for non-euclidean retrieval).
Older Projects & Experiments (Click to expand)
- OSINT: Spanish Blackout with Open Sources.
- Python Wrapper for Spanish Catastro Search.
- RL & Game Solving: (Chess Solver) (Neural Net vs Dummy Agent playing BlackJack).
(Always up for a chess or a poker game!)