Vincent C.

I build AI systems and the applications around them: the data layer, the retrieval and agent logic, the automation, and the web product. My focus is the engineering that makes AI dependable in production: ✦ accurate retrieval + evaluation and critic steps, ✦ human-in-the-loop control, ✦ deployment, including fully on-premise setups with local models. Recent projects: ☞ Multi-agent document intelligence (RAG). A system that answers questions across large document sets with exact page-and-source citations and generates new documents from existing ones. Built on a PostgreSQL vector store with mixed chunking strategies (semantic, per-paragraph, with tables and images handled separately), Hugging Face embeddings, and cosine-similarity reranking. Runs on-premise with local LLMs served through Ollama. ☞ LangGraph agents with custom MCP tools. Agents with tool access over MCP, including servers I implemented where none existed: YouTube control, MinIO/S3-compatible storage, and an internal smart-screen control service driven by natural-language instructions. RAG-based skill retrieval (top-k by semantic similarity) and Redis memory keep behavior consistent across interactions. ☞ AI automation. An n8n workflow that assembles finished PDF manuals from structured product data, and a customer-support email generator backed by a PostgreSQL schema for product/spare-part relationships, with a critic step that auto-sends confident answers and saves uncertain ones as drafts. ☞ Production web platform. A Django application on Google Cloud (Cloud Run, Cloud SQL, GCS, Pub/Sub, Eventarc) with multi-tenant workspaces, resumable uploads, credit billing, REST API access, and admin dashboards. Tools: Python · Django · Streamlit · REST / OpenAPI · LangChain · LangGraph · MCP · n8n · RAG · PostgreSQL / pgvector · Redis · MinIO · Ollama · Hugging Face · Docker · Google Cloud (Cloud Run, Cloud SQL, GCS, Pub/Sub) · HTMX · Tailwind / DaisyUI · Alpine.js