I built CropX, a multilingual agricultural advisory chatbot using FastAPI and a LangChain ReAct agent. It retrieves information from 50,000+ textbook chunks and combines plant-disease image classification with treatment advice.
I also built an agentic RAG system for my MSc dissertation, using FAISS retrieval, query expansion, and multi-hop reasoning. I evaluated retrieval and generation quality, reaching Recall@K above 0.75 and BERTScore above 0.85.
At Ethara AI, I evaluated LLM responses for reasoning accuracy, factual correctness, and instruction adherence. I identified hallucinations and reasoning errors and provided feedback to guide post-training improvements.
