Each of these is a case study rather than a screenshot — what the problem actually was, the architecture I chose, and the parts I'd build differently now.
A face-tracking, talking companion robot on a Jetson Orin Nano. Wake word, speech recognition, retrieval over its own documents, and speech synthesis — all running on-device, with no cloud and no API keys anywhere on the conversation path.
A suite of documentation, research, and development agents sharing context and memory, with model-agnostic routing that picks between local and cloud LLMs based on cost and latency.
A contract analysis and summarization platform that extracts obligations, dates, and risk clauses from legal documents using an NLP pipeline layered under LLM summarization.
A MERN real-time messaging platform with AI-generated smart replies, built alongside a thesis on hybrid credit scoring combining logistic regression with a rule-based system.