ChatBot REX
Lessons-learned records scattered across spreadsheets and email. A chatbot and a KPI dashboard now centralize them for engineers, inside the internal network.
ChatBot REX
Lessons-learned records scattered across spreadsheets and email. A chatbot and a KPI dashboard now centralize them for engineers, inside the internal network.
- Role
- Data & AI developer, end-to-end design and development
- Period
- Sept. 2025 – 2026
- Setting
- SEGULA Technologies
- CamemBERT v3 (fine-tuned)
- ChromaDB
- Sentence-Transformers
- FastAPI
- SQLite
- On-premise GPU inference
In action
Context
SEGULA Technologies engineering department (Montoir-de-Bretagne site): lessons-learned records (REX) were scattered across spreadsheets and email threads, with no single access point. Finding a past lesson relied on human memory, and a departing employee took their knowledge with them. Strong constraint: REX records contain sensitive data that must not leave the internal network.
Mission
Design of a three-layer architecture (API, AI logic, data) with a FastAPI backend and JWT authentication. Implementation of vector search with ChromaDB, backed by a CamemBERT model fine-tuned for semantic search in French. Development of a guided-creation feature: a conversational interview that pre-fills the REX record from the engineer’s free-text answers. To meet the confidentiality constraint, model inference runs entirely on-premise on a dedicated GPU workstation, with no external cloud service involved.
Impact
In production since April 2026, the application is used by engineering department staff: it centralizes REX records, makes them searchable via a chatbot and a KPI dashboard, and supports guided creation directly from the interface. A presentation to the technical management team laid the groundwork for a nationwide rollout.
Let's talk about your project.
Describe what you need in a few lines; we'll see together whether I can help.
Get in touch