NARI AI : AI for Her Health

#devchallenge #googleaichallenge #ai #gemini

NARI AI, is a comprehensive ai powered health platform built for women's health. It operates in two distinct modes:

Clinician Mode: This mode serves as an AI-powered clinical decision support system for medical professionals. It helps doctors and radiologists manage their workflow by automatically triaging patient cases, providing detailed analysis of medical scans, generating draft reports, and offering advanced tools for research and specialized analysis.
Patient Mode: This mode acts as a personal health companion for patients. It aims to empower patients by providing them with simplified, understandable versions of their medical reports, educational resources, and a suite of AI-powered tools to track their health, understand their conditions, and manage their wellness goals.

What problem does this solve?

NARI AI addresses key challenges on both sides of the healthcare experience:

For Clinicians: It tackles issues of workload, diagnostic efficiency, and documentation burnout. By automating the initial analysis of scans and prioritizing cases by urgency, it allows clinicians to focus their expertise where it's needed most. The AI acts as an assistant, drafting reports and providing quick access to research, thereby streamlining the entire diagnostic and administrative workflow.
For Patients: It solves the problem of health literacy and patient disempowerment. Medical terminology and reports are often intimidating and difficult to understand. NARI AI bridges this gap by translating complex information into simple language, allowing patients to ask questions about their results, and providing them with tools to be more proactive and informed about their health journey.

Tech Stack

The application is a modern, single-page web application built with the following technologies:

Frontend Framework: React with TypeScript
UI Component Library: Material-UI (MUI)
Local Database: Dexie.js, which is a powerful wrapper for the browser's IndexedDB. This is used to store patient cases and chat history, enabling offline access and persistence.
AI Integration: The official Google GenAI JavaScript SDK (@google/genai) to communicate with Gemini models.
Client-side Encryption: CryptoJS is used to encrypt data before it's stored in the local database, enhancing privacy.

How does it use the Gemini API?

The application deeply integrates the Gemini API across its features, leveraging its advanced capabilities in multiple ways, as seen in the geminiService.ts file:

Structured JSON Output: It extensively uses Gemini's ability to generate structured JSON output based on a defined schema (responseSchema). This is the backbone of its analytical features, allowing it to turn a medical scan into a detailed object containing risk scores, findings, bounding boxes, recommendations, and more.
Multimodal Analysis (Image + Text Input): For nearly all scan analyses (breast, skin, cephalometric), it sends both the image data and a detailed text prompt to the Gemini model. The model's vision capabilities are used to analyze the image content based on the instructions in the prompt.
Image Generation (imagen-4.0-generate-001): The app uses Gemini to generate new images from text prompts. This is used to create educational illustrations for patients, such as breast self-exam guides, diagrams of fetal development, or visuals for health articles.
Image Editing (gemini-2.5-flash-image-preview): It uses a vision model to perform in-place edits on existing images based on text commands. A user can type "circle the suspicious mass" or "highlight the ribs," and the model returns a new, modified image.
Conversational AI (Chat): It uses the chat API (ai.chats.create) to create specialized AI assistants. These chatbots are given specific system instructions to act as a "patient report explainer" or a "health coach," ensuring their responses are safe, focused, and helpful within their designated role.
Text Generation and Summarization: Standard text generation is used for drafting clinical reports, summarizing research articles, providing second opinions, and creating patient-friendly explanations.

How does it use multimodality?

Multimodality is a core aspect of NARI AI, enabling it to process and generate information across different formats:

Image-to-JSON: The primary diagnostic flow is multimodal. It takes an image (e.g., a mammogram) as input and outputs structured JSON text data that represents the clinical analysis.
Text-to-Image: The educational features are a reverse example. They take a text prompt (e.g., "a diagram of the pelvic floor") and output a new image.
Image & Text-to-Image & Text (Interactive Editing): This is the most advanced use. The app sends an image and a text command (e.g., "color the clavicle blue") to the model. The model processes both inputs and returns a new, edited image along with a text response confirming the action.
Image-in-Chat: The AI Health Plan coach can generate and display images directly within the chat interface in response to a user's text request, making the conversation more visual and engaging.

All Features and Functionalities?

The application is rich with features, divided between the two user modes.

Clinician Features:

Triage Dashboard: View a queue of patient cases automatically sorted by AI-assessed priority (High, Medium, Low).
AI Scan Viewer: An interactive viewer for medical scans with AI findings (masses, lesions) overlaid as color-coded bounding boxes.
Interactive Analysis Chat: A chat interface layered on the Scan Viewer that allows the clinician to give text commands to the AI to edit the image in real-time (e.g., "point an arrow to the calcification").
Advanced Clinical Tools:
- Suggest Differential Diagnoses: Get an AI-generated list of possible diagnoses based on scan findings.
- Draft Clinical Note: Automatically generate a structured clinical note from the analysis.
- Get AI Second Opinion: Request a second AI analysis for confirmation.
- Compare with Prior Scans: An AI-powered tool to summarize changes between a patient's current and previous scans.
Automated Reporting: Generate a detailed clinical report and a simplified patient-friendly report with one click.
Specialized Analysis Modules:
- Cephalometric Analysis: For orthodontists, this tool auto-detects landmarks on an X-ray and provides instant geometric and angular measurements.
- Breast Cancer Analysis: A dedicated view for breast imaging with specialized tools.
AI Research Assistant: A chat interface for asking complex clinical questions and receiving summarized, evidence-based answers.
Education Studio: A tool to generate custom medical illustrations from text prompts for patient education.

Patient Features:

Health Hub Dashboard: A central "bento grid" dashboard giving access to all personal health modules.
My Reports & AI Explainer: View simplified, easy-to-understand versions of your medical reports. Includes an integrated AI chatbot to answer questions about the report in plain language (e.g., "What does BI-RADS 2 mean?").
AI Symptom Checker: Describe symptoms to receive a preliminary analysis of possible conditions and suggested next steps.
Personalized Health Education: An educational portal where you can ask any health question and receive a clear explanation, often with a custom-generated visual aid.
AI Medication Helper: Upload a photo or type the text of a prescription to get a simple explanation of its purpose, dosage, and side effects.
Interactive Scan Annotator: An educational tool to upload a medical scan (like an X-ray) and have an AI label and explain the visible anatomical structures.
Menstrual & Reproductive Health: Track your cycle and symptoms to receive AI-powered insights.
Pregnancy Tracker: Enter your due date to get weekly updates on fetal development, complete with AI-generated illustrations.
AI Health Plan: A conversational tool to chat with an AI health coach, set a wellness goal (e.g., "improve sleep"), and co-create a simple, actionable plan.