.png&w=256&q=75)
1
1
Pakistan
1 year of experience
Software Engineering student with hands-on experience in AI Agents, Agentic AI, Retrieval-Augmented Generation (RAG), and workflow automation. Skilled in Python, C++, Web Development, SEO, and Unity game development. Experienced in building multi-agent AI systems, integrating LLM APIs and tools, implementing RAG with vector databases, and developing AI-powered applications with memory, safety guardrails, and multilingual support. Familiar with n8n automation and API integration. Passionate about hackathons, rapid prototyping, and building practical AI solutions to solve real-world problems.

THE PROBLEM After leaving the hospital, many elderly patients go home with a discharge summary they can't easily read or remember. They forget doses, don't know who to ask, and families can't tell if the plan is being followed. Typing is hard for them, and general chatbots may guess or give unsafe medical advice. OUR SOLUTION Sehat Saathi ("Health Companion") is a voice-first web app with patient and caretaker sides. Patient app: - One big microphone button: ask a question, get a spoken and written answer. - Every answer shows the numbered line of the discharge summary it came from, plus a confidence label (Certain, Possibly, Unknown). - If the answer isn't in the summary, the app says so and adds the question to a review queue for family or doctor. It never diagnoses or advises doses. - Before saving a reminder or booking a visit, it repeats what it heard and waits for a big Yes or No. - Full-screen medicine reminders and mock appointment booking. - SOS button and emergency contacts. A fixed safety filter spots red-flag phrases such as chest pain, shows an emergency screen, tells the patient to call the emergency number, and alerts family. No medical advice. Caretaker app: - Dashboard of open alerts, review items, reminders taken or missed, and the next visit. - Alerts show the patient's exact words, with a "handled" action. - Private pairing: the patient shows a one-time key valid for 10 minutes and must approve the link. They can remove it any time. SAFETY The red-flag check runs before the language model and overrides it. Answers are grounded in the patient's own summary. Demo data is synthetic. This is a prototype, not a replacement for a doctor or emergency services. BUILT WITH Mobile-first web app designed in Figma, with a FastAPI backend, speech-to-text, and a language model served with vLLM. [Add your AMD details here.
18 Oct 2026