Doc Dolphin

Live Demo: doc-dolphin.netlify.app

Project board: Trello

Doc Dolphin is an AI-powered health companion that analyzes lab-report images, tracks daily mood, and generates insights about your emotional well-being. A multimodal vision model reads the report image directly, and its analysis is streamed back to the browser token-by-token.

Medical disclaimer: Doc Dolphin is not a substitute for professional medical advice, diagnosis, or treatment. Always seek the advice of your physician or a qualified mental health provider.

Features

Test Analyzer

• Upload up to 10 lab-report images (JPG/PNG); they are merged and preprocessed before analysis
• AI-generated, streaming explanation of the results, rendered as live Markdown
• Save analyzed reports (with an auto-generated title) to your history

Results Dashboard

• Centralized, paginated view of all saved reports
• Open any past report to revisit its analysis

Mood Journal

• Daily mood logging with weather, activities, and free-text notes

Mood History

• Sortable table of past entries with inline edit and delete

Mood Insights

• AI analysis of mood entries over a chosen date range, streamed live
• Personalized observations and tips based on your journal

Architecture

flowchart LR
    subgraph Client["React + Vite client"]
        UI["Upload / Mood UI"]
        Stream["fetch + TextDecoderStream reader"]
    end
    subgraph Server["Express API"]
        Auth["JWT auth middleware"]
        Pre["sharp image preprocessing"]
        SSE["PassThrough → text/event-stream"]
    end
    Ollama["Ollama vision LLM\n(llava-llama3)"]
    Mongo[("MongoDB / Mongoose")]

    UI -->|"FormData (multer, 10MB, jpg/png)"| Auth
    Auth --> Pre --> Ollama
    Ollama -->|streamed tokens| SSE --> Stream --> UI
    Auth -->|reports, moods, users| Mongo

Loading

How it works

• Real token streaming, end to end. The server runs Ollama with stream: true, pipes each chunk through a Node PassThrough as text/event-stream, and the client consumes it incrementally with response.body.pipeThrough(new TextDecoderStream()).getReader() plus an AbortController stop button. No polling, no fake typing.
• Vision-LLM document understanding (no separate OCR step). Uploaded images are rotated, resized to a fixed width, grayscaled, and vertically composited into a single optimized JPEG with sharp, then passed base64 to a multimodal model (llava-llama3 via Ollama). This cuts token cost and latency while letting the model read the report directly.
• JWT auth with bcrypt. Passwords are hashed with bcrypt and validated with validator (strong-password + email checks). Login issues a short-lived (3d) JWT; an Express middleware verifies the Bearer token on every protected route.
• Server-side pagination. Saved reports are paged with a single Mongoose $facet aggregation that returns the page data, total count, and total pages in one round-trip.

Tech Stack

Frontend

• React (Hooks + Context) with Vite
• shadcn/ui + Tailwind CSS + Lucide icons
• React Router
• Fetch API with the Streams API for live token streaming

Backend

• Node.js + Express 5
• MongoDB + Mongoose 8
• bcrypt for password hashing, validator for input checks
• JWT for authentication
• multer for image uploads, sharp for preprocessing

AI

• Ollama running a llava-llama3 multimodal model
• Vision-based lab-report reading (model reads the image directly)
• Streaming analysis for both lab reports and mood insights

Screenshots

Light and dark captures of the main screens. (Images live in client/public/.)

Analyzer	Results

Log Mood	Mood History	Mood Insights

Prerequisites

• Node.js 18+
• MongoDB (local or a connection string)
• Ollama running locally, with the vision model pulled:

  ollama pull llava-llama3:8b-v1.1-fp16

Installation

1. Clone the repository

git clone https://github.com/Rajskij/doc-dolphin.git
cd doc-dolphin

2. Server (Express API)

cd server
npm install
cp .env.example .env   # then fill in the values below
npm start

3. Client (React + Vite)

cd client
npm install
cp .env.example .env   # set VITE_API_URL to your running server
npm run dev

Configuration

Backend — server/.env

Variable	Required	Description
MONGO_URI	yes	MongoDB connection string
JWT_SECRET	yes	Secret used to sign/verify JWTs
PORT	yes	Port the API listens on (e.g. 8000)
OLLAMA_URL	yes	Base URL of your Ollama server (e.g. http://localhost:11434)

Frontend — client/.env

Variable	Required	Description
VITE_API_URL	yes	Base URL of the Express API (e.g. http://localhost:8000)

License

MIT