AI / MACHINE LEARNING ENGINEER
Auto Grading API
Auto Grading API v1 is a Retrieval-Augmented Generation system that evaluates student assignment documents against JSON-based rubrics. The FastAPI service extracts text and metadata from PDF files, retrieves relevant content with SentenceTransformer and FAISS, and uses Gemini or OpenRouter to produce structured grading results.
View on GitHub
My Contribution
- Developed the FastAPI service for automated grading of student PDF assignments.
- Built PDF text and metadata extraction for uploaded assignment documents.
- Implemented semantic search with SentenceTransformer embeddings and a FAISS index.
- Developed the RAG grading pipeline to retrieve relevant assignment context before evaluation.
- Configured chunk size, chunk overlap, embedding model, and top-k retrieval through environment variables.
- Implemented JSON-based rubric processing for structured and consistent evaluation criteria.
- Integrated Gemini and OpenRouter LLM APIs with configurable model selection and API keys.
- Exposed the POST /api/v1/grade endpoint for rubric and assignment uploads.
- Structured API responses with assignment metadata and advancedgradingdata results.
- Added centralized logging and a batch CLI mode for grading multiple PDF files.
- Configured the project for local development with Uvicorn and interactive API documentation.