An open-source, anonymous mental-health companion — a crew of specialist agents reads every message in parallel, deliberates, and a critic reviews the reply before you ever see it. It knows when to hand off to a real psychologist.
A college runs its whole certification programme on one app — students submit certificates, mentors approve them, and an approval turns into real internal-exam marks or an elective exemption. Nine roles, one login, and every query scoped to the department you actually belong to.
A voice-driven AI scribe for radiologists — dictate your findings and it returns a clean, structured report. It never reads images; the doctor reviews and signs off every word.
An adaptive multi-tier LLM cascade router. Every query starts at the cheapest model that could plausibly answer it and escalates only when a judge confirms the cheap answer actually failed — measured at 92.7% cost saved, on entirely free-tier APIs.
A RAG-powered medical Q&A chatbot — it grounds every answer in real medical documents instead of hallucinating. This is where I first learned RAG and LangChain.
A life-management platform — habits, journal, tasks and an AI assistant that can act on all three by chat. Built as an experiment in how far AI-assisted 'vibe coding' actually goes: 15+ Prisma models, tRPC end to end, and journal entries encrypted at rest.
A voice-command system that only obeys one registered speaker — built during my first internship at Equidor Medtech. It verifies who is talking before it listens to what they said, and runs entirely offline.