Baptiste GASTON
Back to CV

Studies

AI Core

Local AI demonstrator with three modes: chat, RAG, and agent.

Academic project - Since February 2026

Tech stack

Node.js TypeScript Ollama Express Prisma MySQL Docker

Project background

Project commissioned by Synertic as part of the SAE S5.01 coursework. The goal was to build a demonstrator for testing LLM integration within an existing chat application. The deliverable: a single web page where users can type a message, select a response mode, and get an automatically generated response from a local AI model. Built in a team of six students.

Project highlights

  • Three modes: classic chat, RAG with document input, and agent mode with action planning.
  • Locally hosted LLM via Ollama, integrated into a Node.js/TypeScript service.
  • Docker architecture with Express, Prisma, MySQL, and REST endpoints.
  • Full processing traceability: execution IDs and per-step summaries.

Three operating modes

The demonstrator exposes three distinct modes from a single interface. Classic chat mode sends messages directly to the LLM. RAG mode lets users inject documents, which are chunked and indexed to enrich the response context. Agent mode plans a sequence of actions to execute using a tool-calling mechanism.

Technical architecture

Node.js/TypeScript service exposed via Express, with a data model managed by Prisma on MySQL. The LLM is hosted locally via Ollama, with a European-origin constraint set by Synertic. The whole stack is deployed via Docker Compose. Every request is traced: execution IDs, per-step summaries, and a log of tools called.