How sampreethg/-intelligent-university-knowledge-assistant works
UniMind is a Streamlit-based academic knowledge assistant. Its main workflow accepts institutional documents, parses and chunks them, indexes their embeddings alongside SQLite metadata, then retrieves relevant passages to generate citation-grounded answers with local Ollama inference. The graph also includes document inspection, answer telemetry/export, and optional remote demonstration. The README documents the overall flow; sampled code confirms parsing, chunking, storage, and grounded generation behavior, but does not show every query-path call site.
Main entry points: app.py Streamlit Portal · vector_store.py Similarity Retrieval · llm.py Grounded Prompting · ingestion.py Document Parser
Components 13
User Experience
- Streamlit Portal
app.py - Chunk Inspector
app.py - Citations & Export
app.py - Cloudflare Tunnel
Document Ingestion
- Document Parser
ingestion.py - Overlap Chunking
chunking.py
Knowledge Retrieval
- Text Embeddings
vector_store.py - FAISS Vector Index
vector_store.py - SQLite Metadata
database.py - Similarity Retrieval
vector_store.py
Grounded Answering
- Grounded Prompting
llm.py - Local Ollama
Other components
- Student or Faculty
Connections 18
- Student or Faculty → Streamlit Portal: uploads or asks
README.md - Streamlit Portal → Document Parser: parses uploads
README.md - Document Parser → Overlap Chunking: returns pages
app.py - Streamlit Portal → Overlap Chunking: chunks pages
app.py - Streamlit Portal → Text Embeddings: indexes chunks
README.md - Text Embeddings → FAISS Vector Index: adds vectors
vector_store.py - Text Embeddings → SQLite Metadata: records chunks
vector_store.py - Streamlit Portal → Similarity Retrieval: searches query
README.md - Similarity Retrieval → Text Embeddings: embeds query
vector_store.py - Similarity Retrieval → FAISS Vector Index: searches vectors
vector_store.py - Similarity Retrieval → SQLite Metadata: loads metadata
vector_store.py - Streamlit Portal → Grounded Prompting: requests answer
README.md - Grounded Prompting → Local Ollama: streams inference
llm.py - Similarity Retrieval → Grounded Prompting: provides passages
README.md - Grounded Prompting → Citations & Export: streams cited answer
app.py - Streamlit Portal → Chunk Inspector: opens chunks
README.md - Student or Faculty → Citations & Export: views or exports
README.md - Student or Faculty → Cloudflare Tunnel: shares remotely
README.md