Benjamin Houdebert

Designed and implemented a dynamic Retrieval-Augmented Generation application connected to multiple LLM providers.

The goal was to create a reusable knowledge layer capable of ingesting, vectorizing, indexing, and pre-warming internal documentation. The application uses semantic search, vector distance, contextual retrieval, and LLM orchestration to provide reliable, context-aware answers to client requests while remaining adaptable to different business domains.

Stack:

- Python

- Rust

- ChromaDB or Postgres PGVector

- local LLM model for encoding

- External LLM for reasoning (openai, anthropic ...)

Built a consumer web platform for mountain bike enthusiasts to centralize, compare, and manage bike geometry data.

The application helps riders make better fit and purchase decisions by providing structured geometry management, SVG-based bike visualizations, side-by-side comparisons, and a personal bike collection synchronized with a shared library. It also supports data import/export, responsive PWA usage, templating, versioning, diff and synchronization workflows, rider profile tuning, community voting, and scheduled calendar-based events.

Stack:

- ​React TypeScript (Redux Toolkit, TailwindCSS, Vite PWA) frontend

- Rust + Actix‑web backend with SQLX;

- PostgreSQL database; Docker Compose and Nix

Technical leader on a large-scale Delivery Management System used across multiple business units:

Stack: