Back to castle

Potions

The Magic Behind the Sorting Hat

Skills

  • LLM integration
  • Retrieval-Augmented Generation
  • Vector search
  • Prompt engineering
  • Full-stack
  • PostgreSQL
  • Caching

How the Chatbot Works

The RAG Pipeline

  1. 01

    Embed

    Résumé, project, and profile content is chunked and embedded by an embedding model.

  2. 02

    Index

    The embeddings are stored in a local index file — no external vector database.

  3. 03

    Retrieve

    A question is embedded and matched to the closest passages by cosine similarity.

  4. 04

    Answer

    An LLM replies using only the retrieved passages, never inventing facts.

  5. 05

    Protect

    A response cache and per-visitor rate limiting guard every request, keeping the chatbot fast and free to run.