Sergio Márquez

Backend engineer applying AI to real products and internal tooling.

• Automated RAG ingestion via n8n + Python (Sheets -> MongoDB + Pinecone) with hash dedupe; removed manual uploads; initial loads of 500-600 docs; saved ~8-10 hours per setup.
• Improved document validation pass rate from 70% to 90% using a GPT+Claude+Gemini ensemble, prompt/parse improvements, and OCR tuning (Tesseract, OpenCV).
• Migrated chat to an agentic architecture (Google ADK) with LiteLLM; retrieval on MongoDB with custom chunking and reranking; ~15% lower median latency and ~35% lower infra cost.

Owned the Java/Spring backend and supported cloud/data integrations.

• Middleware APIs with Spring connecting Angular frontends and third-party services.
• Workloads on Kubernetes with automated GitLab CI/CD pipelines.
• Post-merger data integration (Preving/Cualtis): analysis, mapping, and refactors.