Allom logo

Allom

Matching platform connecting businesses with on-demand experts. Press coverage in Le Point and Entreprendre.

Allom platform

Date:

View live

Role: Founder & CTO

AI Matching NLP Marketplace Vector Search Full-stack

What it is

Expert matching platform that connects businesses with on-demand consultants. AI-powered: users describe their problem in natural language, the system returns ranked expert profiles in seconds.

Matching engine

The core differentiator. Built a semantic matching pipeline from scratch:

  • FAISS vector index for sub-millisecond similarity search
  • CamemBERT / FlauBERT for French-language embedding generation
  • Cosine similarity scoring between user queries and expert profile vectors
  • Re-ranking layer combining semantic score, availability, and rating

Architecture

  • Frontend: Next.js + Chakra UI + Redux. Custom “atomic blocks” component architecture for granular reusability.
  • Backend: FastAPI (async Python). Migrated from Django in alpha. Piccolo ORM.
  • Infrastructure: AWS SageMaker (ML models), S3 (FAISS indices), Heroku (API + Redis), Vercel (frontend)
  • Data: PostgreSQL + Redis (session management, real-time messaging buffer)

Press

Website: www.allom.fr