Physics Word Problem Generation

Generate high-quality, diverse physics word problems using LLMs and agentic Retrieval-Augmented Generation (RAG) with Qdrant vector database.

Folder Structure

• DATASET: Saved physics questions by topic.
• ENTITY: Topic entities with words and properties.
• FINE_TUNE_GPT2: GPT-2 fine tuning code.
• LLM: RAG chains for question generation.
• LLM_CONFIG: LLM selection/configuration.
  • BUILD_DATASET: Save generated questions if true.
  • OPENAI_API_KEY: OpenAI API key.
  • INNER_MODEL: OpenAI model name.
• NOVELTY_SCORE: Novelty score calculation.
• TOPICS: Topic equations.
• UTILS: Utility functions.

Agentic RAG with Qdrant

This project uses an agentic RAG pipeline to iteratively retrieve and refine context chunks from a Qdrant vector database. The agent interacts with Qdrant, evaluates chunk relevance, and improves retrieval over multiple steps to maximize question quality and topic coverage.

• Qdrant: Stores vectorized textbook and question bank chunks.
• Agentic RAG: The agent queries Qdrant, analyzes retrieved chunks, and adapts its strategy to get better context for question generation.
• Iterative Retrieval: Each round, the agent refines its queries and chunk selection, leading to more accurate and novel physics questions.

How to Run

physics.py is the main entry point. Currently we have added the following topics.(see picture below)

python physics.py

Citation

If you use this codebase, please cite or acknowledge appropriately. For questions or suggestions, open an issue or contact the maintainer.

---