Context
The problem
Turn instructor course PDFs into grounded, exam-ready questions with answer keys.
System
Architecture
- Course PDF parsing
- Embeddings and semantic retrieval
- PostgreSQL with pgvector
- FastAPI backend
- Pydantic schemas and structured LLM outputs
Engineering judgment
Technical decisions
Index course content for semantic retrieval
The ingestion pipeline parses course PDFs, generates embeddings, and indexes content in PostgreSQL/pgvector.
Use reusable question-family templates
The workflow converts generated questions into reusable templates and instantiates student-specific variants.
Validate generated outputs
The verification service checks answer correctness, concept consistency, and source provenance through typed Pydantic schemas and structured LLM outputs.
Result
Outcome
Contributed ingestion, reusable question-family, and verification capabilities to the co-developed assessment platform.