Projects
- Graph-based Planning System for LLMsA graph neural network that retrieves structured plans to improve LLM reasoning on multi-hop question answering.
- Closed-Domain Multimodal QAFine-tuned vision-language models with LoRA to answer questions over technical graphs and tables, +12.4% on multi-hop QA.
- RAG Question Answering with LangChainA retrieval-augmented QA system using FLAN-T5 and FAISS vector search, orchestrated with LangChain.