Building production AI systems that actually work — not demos. I specialize in RAG pipelines, autonomous agents, and security auditing.
AI/ML: RAG systems, hybrid retrieval, LLM evaluation, autonomous agents
Backend: Python, FastAPI, PostgreSQL, Redis, Docker
Frontend: React, Next.js, TypeScript
Security: Network auditing, JWT analysis, threat hunting (KQL)
| Project | What It Does | Status |
|---|---|---|
| rag-demo | Production RAG: hybrid search (BM25 + embeddings + RRF + rerank), multi-tenancy, autonomous agents. Deploy in 1 click. | ✅ Live |
| hybrid-rag-evals | RAG híbrido: BM25 + embeddings + RRF + rerank. Recall@5: 0.643 → 0.964. 93.27% coverage. | ✅ Live |
| medusa | Autonomous second brain: 6,000+ chunks RAG, 6 agents, voice, auto-publishing. | ✅ Running 24/7 |
| security-toolkit | Security tools from scratch (stdlib only): portscan, subenum, httpaudit, jwtdump. | ✅ 16/16 tests |
| tryhackme-roadmap | 90-day security learning plan: THM → HTB → KC7 → certs. | 📋 In Progress |
- Measure everything: hybrid-rag-evals has a full evaluation harness that shows exactly what improves and where it fails.
- Write real code: No wrappers, no black boxes. I implement BM25, RRF, and cross-encoders from scratch to understand them.
- Document failures: The write-ups I write focus on dead ends, not just flags.
- Advanced RAG: pgvector HNSW, semantic chunking, production evaluation pipelines
- Agentic Systems: LangGraph state machines, human-in-the-loop, memory management
- Security: Threat hunting with KQL, CTF challenges, vulnerability research
- Email: paz243482@gmail.com
- LinkedIn: miguel-hernandez-04129542a
- GitHub: MaikolProx
- TryHackMe: paz243482
- KC7 Cyber: profile
Building systems that work. Measuring everything. Sharing what I learn.