| --- |
| title: AidAILine |
| emoji: 🩺 |
| colorFrom: blue |
| colorTo: green |
| sdk: gradio |
| sdk_version: 5.49.1 |
| app_file: app.py |
| pinned: false |
| license: mit |
| short_description: Local-first medical record keeper for caregivers |
| tags: |
| - build-small |
| - track:backyard-ai |
| - badge:off-brand |
| - badge:best-demo |
| --- |
| |
| # 🏥 AidAILine |
|
|
| > **Build Small Hackathon — Backyard AI Track** |
| > Built for caregivers managing multi-provider care for a dependent. Local-first. AI-powered. Nothing leaves your computer. |
|
|
| ## Demo & Social Links |
| - **Live Space:** https://huggingface.co/spaces/build-small-hackathon/AidAILine |
| - **Demo Video:** _(upload a screen recording to YouTube as unlisted, then paste the URL here)_ |
| - **Social Media Post:** https://x.com/JaccobBarrett/status/2066667333943205985 |
|
|
| --- |
|
|
| ## The Story |
|
|
| Caregivers track a lot. Doctors, physical therapy, weekly prescriptions, monthly refills, lab results from three different clinics, allergies, and the questions you wish you'd asked at the last visit. |
|
|
| Today, that lives on a paper calendar with sticky notes pinned to it. Things fall through the cracks. |
|
|
| **AidAILine replaces the paper calendar and the sticky notes** with a local-first Gradio app that runs on your own hardware. A small AI model (Qwen 2.5 7B Instruct via the HF Inference API) reads your uploaded medical PDFs and answers questions about them in plain language. The app itself is small enough to read end-to-end. Your data never leaves the app. |
|
|
| No telemetry. No analytics. No cloud sync. No accounts. |
|
|
| --- |
|
|
| ## What It Does |
|
|
| | Tab | What it does | |
| |---|---| |
| | 📄 **Documents** | Upload PDFs → ask questions in natural language via small AI | |
| | 💊 **Medications** | Track current & past meds with dosage, frequency, side effects, and notes | |
| | 📅 **Appointments** | Upcoming appointments with timeline, prep notes, and a calendar-aware add/edit flow | |
| | 🍽️ **Food Chart** | Liked / Disliked / Allergy foods in three columns — print and bring to the visit | |
| | 📋 **Doctor Visit** | Pre-Visit Brief generator — pulls meds, allergies, and your custom questions into a print-ready handout | |
| | 📤 **Export & Share** | One-click copy, mailto link, or print for the full caregiver summary | |
| | ⚙️ **Settings** | Patient info, model path, privacy-first portal integrations (clearly marked as future work) | |
|
|
| **The pre-visit brief is the killer feature.** Walk into grandma's appointment with a single page that says: "Here's her current meds, here are the foods she can't eat, here are the three questions I want to ask today." Most caregivers bring nothing. AidAILine gives you a reason to bring everything. |
|
|
| --- |
|
|
| ## Tech Stack |
|
|
| | Component | Library | |
| |---|---| |
| | UI | Gradio 5 | |
| | AI (Documents tab) | Qwen 2.5 7B Instruct via Hugging Face Inference API | |
| | Embeddings | sentence-transformers (`all-MiniLM-L6-v2`) | |
| | Vector Store | FAISS | |
| | PDF Parsing | PyMuPDF | |
| | Storage | Local JSON files | |
|
|
| All models are under 32B parameters. The app runs on a free Hugging Face CPU Space. |
|
|
| --- |
|
|
| ## Why "Backyard AI" |
|
|
| This is the textbook backyard build: |
|
|
| - **Small model** — Qwen 2.5 7B runs on the HF Inference API, free tier |
| - **Local storage** — everything is plain JSON in `data/` |
| - **Local PDF parsing** — no cloud OCR |
| - **No telemetry** — open the network tab, watch nothing happen |
| - **No accounts** — no signup, no login, no password reset |
| - **Open weights** — Qwen 2.5 is Apache 2.0 |
|
|
| The 6 core tabs (medications, appointments, food, doctor visit, export, settings) run fully offline. Only the Documents tab calls the Inference API, and only to run the small model — your data and your prompts are not stored or logged by the model host. |
|
|
| --- |
|
|
| ## Privacy |
|
|
| The 6 core tabs are 100% offline. All patient data is stored as plain JSON files in the `data/` directory. No analytics, no telemetry, no accounts. The Documents tab uses a hosted model, but only the question you type is sent to the model — the model does not store or train on your prompts. |
|
|
| --- |
|
|
| ## Project Structure |
|
|
| ``` |
| build-small/ |
| ├── app.py # Gradio entry point |
| ├── rag_engine.py # PDF RAG pipeline (embed → FAISS → hosted Qwen) |
| ├── pdf_loader.py # PDF parsing + chunking |
| ├── med_tracker.py # Medication CRUD |
| ├── appointment_tracker.py # Appointment CRUD |
| ├── food_tracker.py # Food chart CRUD |
| ├── profiles.py # Multi-profile support |
| ├── doc_forms.py # Pre-Visit Brief + Doctor Visit generator |
| ├── export_tool.py # Export + mailto builder |
| ├── config.py # Paths, settings |
| ├── requirements.txt |
| ├── data/ # Patient data (JSON) |
| └── assets/ |
| ``` |
|
|
| --- |
|
|
| ## What's Next (v2) |
|
|
| Captured in a separate vault, planned for after the hackathon: |
|
|
| - Manager / Sub-Profile architecture — primary account manages care for multiple dependents |
| - Medication refill & prescription renewal reminders |
| - Types of Care provider directory — categorized directory with frequency tracking |
| - One-tap phone calls to pharmacy and prescriber |
|
|
| The v1 ships the foundation. V2 makes it a household medical command center. |
|
|
| --- |
|
|
| ## License |
|
|
| MIT |
|
|
| *Built for the Build Small Hackathon — Backyard AI Track* |
|
|
