AI Agents Projects

Build a retrieval-augmented generation (RAG) pipeline end to end. Chunk a fictional company handbook the model has never seen, embed it with MiniLM (mean pooling + L2 normalisation, no sentence-transformers needed), explore the embedding space, and measure retrieval recall@k. Then ground Qwen2.5-1.5B-Instruct's answers in the retrieved passages and show that retrieval turns hallucinated guesses into correct answers. The first run downloads about 3 GB of models.

Build a tool-using agent from scratch around a small open LLM (Qwen2.5-1.5B-Instruct). Write three tools (an AST-based safe calculator, a knowledge-base lookup and a unit converter), a ReAct prompt, a robust parser that turns malformed model output into error observations, and the Thought → Action → Observation loop with a max-steps guard. Then score the agent on ten tasks, compare it with answering directly, and chart every trace. The first run downloads the model (about 3 GB).