All AI systems
Case study · Gwee Per Ming
INTI-MBA RAG Chatbot
A retrieval-augmented study assistant that answers from course materials with citations and keeps its indexed corpus current through scheduled ingestion.
The problem
Students need answers grounded in their course documents, while newly added or changed Canvas materials should not require repeated manual uploads.
My contribution
Engineered a LangGraph RAG agent with multi-query retrieval, document grading, and query-rewriting self-correction. Added daily Canvas ingestion and OCR for scanned or image-based documents.
Architecture and stack
FastAPI, LangGraph, LangChain, hybrid retrieval, Pinecone, RapidOCR, APScheduler, AWS EC2, and Next.js.
- FastAPI
- LangGraph
- Pinecone
- RapidOCR
- APScheduler
- AWS EC2
- Next.js
What runs today
Used across four live Canvas LMS courses. Scheduled ingestion keeps the indexed materials current without manual uploads.
Outcome
Maintains 215+ documents and 3,800+ searchable passages, returning cited answers grounded in source files.
Proof and source links
The public demo or repository link will be added after a cleaned, shareable version is available.