All posts
aivector-searchragreact

What I learned building Concious

An overview of building Concious: a full-stack second brain application integrating hybrid semantic search and a RAG chatbot using MongoDB Vector Search.

Personal knowledge management tools often fail when search is limited to exact keyword matching. With Concious, I built a second-brain platform designed around hybrid lexical-semantic retrieval.

The Retrieval Pipeline

To balance exact matches (like names or specific URLs) with conceptual matches, the search runs two parallel processes:

  1. Lexical Retrieval: Direct text index queries mapping keywords.
  2. Semantic Retrieval: Generating high-dimensional vector embeddings of saved links and articles using the Hugging Face API, then querying a MongoDB Atlas Vector Search index.

The RAG chatbot (Ashqnor) pulls context directly from these matched nodes to ensure generated responses contain precise source citations.