For Proximus, 2025BrandBot
Proximus's internal AI assistant: brand guidelines, logos, visuals and tone of voice, in one question.
- Brand book
- Tone of voice
For Proximus, 2025- Client
- Proximus
- Role
- AI architect
- Platforms
- Web
- Stack
- Python, FastAPI, Next.js, LangGraph, Azure, Bynder API
- Year
- 2025
The context
Proximus, Belgium's leading telecom company, wanted its teams to instantly find brand rules, logos, official visuals and the right tone of voice. The job was to take a prototype to production.
What I did
- RAG pipeline connected to the company DAM (Bynder): logos, images and PDF guidelines
- Automatic request classification, by brand and question type
- Specialized agents: guidelines, asset search, tone of voice, query rewriting
- Tuned chunking, rewriting and filtering for accurate answers
- FastAPI backend orchestrating the AI flows, and an internal web interface
- Migration to Azure: Azure Vector Search, Azure PostgreSQL, production release
The result
The assistant went from prototype to production on Proximus's Azure infrastructure, with full documentation and training so internal teams could take over.
In pictures
Next projectCapsule
