L

LLM Application

🚑 MedicAssist – Tactical Field Medic Assistant

MedicAssist is an offline, CPU-first Progressive Web App (PWA) that transforms unstructured medical audio or text into structured FHIR R4 JSON. Designed for first responders, disaster relief teams, and combat medics, it enables fast and standardized patient documentation without internet connectivity.

Using Whisper.cpp for offline speech-to-text and Phi-3 Mini with llama.cpp for clinical information extraction, MedicAssist generates valid FHIR R4 Bundles containing patient details, injuries, treatments, observations, and triage information. All processing runs locally on the CPU, ensuring privacy, reliability, and uninterrupted operation in remote or emergency environments.

Key Features 🎤 Offline speech-to-text with Whisper.cpp 🤖 CPU-only AI extraction using Phi-3 Mini 📋 Converts unstructured input into FHIR R4 JSON 🌐 Progressive Web App with full offline support 💾 Local storage using IndexedDB 📤 Export FHIR JSON for later synchronization 🔒 Privacy-first with no cloud or internet dependency

MedicAssist demonstrates how offline AI can convert unstructured medical data into standardized healthcare records, making it ideal for emergency response and disaster management.

F2P Solver extracts mathematical formulas from PDF research papers and converts them into interactive Python simulations — entirely offline, on CPU. Upload a paper, and for every formula you get a callable Python function plus a live chart with sliders for each variable: drag a slider and the curve updates instantly, with no further AI calls. Built with Streamlit, pymupdf, ollama (Qwen2.5-7B), numpy, and Plotly. Runs on any Linux/macOS machine with a single command: ./run.sh.