README.md

metadata

title: Plutus Learning Recommender
emoji: 📚
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false

Plutus Learning Material Recommender

Simple content-based recommendation system built on the Plutus Learner RAG corpus.

Features

• BM25 ranking (pure Python, zero heavy dependencies)
• Query-based recommendations
• Topic-based recommendations
• Filter by tool (plinth / cardano)
• Runs on Colab CPU and free Hugging Face CPU Spaces

Local / Colab usage

# In the plutus_recommender folder
python recommender.py "What is the difference between datum and redeemer?"

Or in a notebook:

from recommender import PlutusRecommender
rec = PlutusRecommender()
results = rec.recommend("Plutus V3 validators and ScriptContext", k=5)
for r in results:
    print(r["title"], r["score"], r["url"])

Deploy to Hugging Face Space (CPU)

1. Create a new Gradio Space on Hugging Face.
2. Upload the entire plutus_recommender/ folder contents:
   • app.py
   • recommender.py
   • rag.py
   • data/ (the index folder)
   • requirements.txt
   • README.md (this file)
3. The Space will start automatically. No GPU needed.

Optional: Dense / Hybrid mode

For semantic recommendations install FastEmbed (still CPU):

pip install fastembed numpy

Then in code:

# First build vectors (one-time, in Colab)
from rag import embed_index
embed_index("data/index")

# Then
rec = PlutusRecommender(mode="hybrid")

Corpus note

The bundled index contains 15 bootstrap summary chunks.
To expand the corpus, use the original ingest.py from the RAG bundle against sources.json.