UniBot

Local RAG agent for answering questions about TED University regulations, courses, and internships.

Stack

• Ollama with gemma3:4b
• ChromaDB
• all-MiniLM-L6-v2 embeddings
• LangChain

Flow

TEDU source data → cleaning and chunking → embeddings → ChromaDB retrieval → Gemma 3 response.

Setup

Requires Python 3.10+ and Ollama.

ollama pull gemma3:4b
python -m venv venv
source venv/bin/activate       # Windows: .\venv\Scripts\activate
pip install -r requirements.txt
python reset_brain.py
python rag.py

The included tedu_clean_data_final.txt can be used to build the vector store without scraping again.

To refresh the source data, run map_links.py, mass_ingest.py, cleaner.py, then reset_brain.py.