🛰️
AI and research
Public

Veille IA

An AI watch platform. It gathers in one place what AI labs and vendors publish — OpenAI, Anthropic, Google DeepMind, Mistral, Hugging Face, NVIDIA, Microsoft — along with a few specialised outlets.

The idea came from my own routine: keeping up with AI meant opening fifteen blogs every morning. The application does it for me, once a day.

What I built

  • Automatic collection from 16 sources: RSS, Atom and RDF feeds parsed by hand, metadata extracted for sites without a feed.
  • Three-pass deduplication: canonical URL, feed identifier, title + date fingerprint.
  • Full-text search in MySQL boolean mode, with a fallback for short terms such as "AI".
  • Personal analysis notes attached to articles.
  • Cron-driven synchronisation and manual runs queued from the admin panel.
  • Respect for the sources: robots.txt, conditional requests, excerpts only and a link to the original.

Written in plain PHP 8.4, no framework: router, dependency injection, templating, cache and logging are hand-written; HTMLPurifier is the only production dependency. Strict content security policy, CSRF tokens, Argon2id, SSRF guard, and analytics that only start after consent.

Role: design, full stack development and production deployment, from the data model to the server. 88 automated tests.

Skills in practice

Automate periodic collection and make information searchable through full-text search in a PHP/MySQL application.