All projects
02 · Campus AI

Ask-Assist

An AI-powered campus assistant combining hybrid intent classification, structured lookups, and retrieval over institutional PDFs.

ask-assist.log
query → intent classification
PDFs → ChromaDB retrieval
contacts / locations → SQLite
Mistral-7B → formatted response

Built a campus assistant with FastAPI and Streamlit, using a three-level hybrid intent classification pipeline to route queries.

Implemented a RAG pipeline with ChromaDB and Sentence Transformers using 384-dimensional vectors, grounding responses in institutional PDF documents.

Designed a SQLite schema and fuzzy entity matching for contact and location lookups. Integrated Mistral-7B for response formatting with confidence-based fallback handling.

Highlights

  • Three-level hybrid query routing
  • Context-grounded PDF retrieval
  • Structured lookups with confidence-based fallbacks
Next projectGLUG App