HN Showcase
Back to browse
GitHub Repository

FastAPI backend for brush-up-py deployed on Fly.io

3 starsPython

AI tutors for Python, Ruby, and JavaScript powered by my personal notes

by jow13_12·Apr 29, 2026·1 point·0 comments
Visit ProjectView on HN

AI Analysis

●●SolidBig BrainNiche Gem

Graph traversal over Zettelkasten notes beats naive vector search for code tutoring.

Strengths
  • •TF-IDF retrieval plus 1-hop graph expansion keeps context bounded and relevant.
  • •Obsidian-style wikilinks turn static notes into a queryable knowledge graph.
  • •In-memory token budgeting prevents abuse on public demos without complex infra.
Weaknesses
  • •Limited to three languages and relies entirely on author's personal note corpus.
  • •No persistence layer for user conversations or long-term learning history.
Category
AI/ML
Target Audience

Developers building RAG systems or personalized AI tutors

Similar To

Obsidian · Logseq · Cursor

Post Description

Live demos (also linked at the top of the repo):

- https://joecardoso.dev/brush-up-py

- https://joecardoso.dev/brush-up-rb

- https://joecardoso.dev/brush-up-js

The retrieval engine, although simple, sparks some interesting ideas, e.g.: what if instead of Zettelkasten notes you had summaries of codebase sections, and instead of [[wikilinks]] you had signposts to architectural dependencies? Graph traversal over a codebase map could be a way to help agents navigate large repos while keeping context window usage deliberate and bounded.

Similar Projects