PersonalDocAI

TL;DR

PersonalDocAI is an intelligent assistant for individual therapy that uses Retrieval Augmented Generation (RAG) to provide personalized consultations based on your medical data. The application integrates test results, medical documents, and observation history to deliver meaningful recommendations and support health-related conversations. All data is processed locally, and interaction occurs through a user-friendly web interface.

Usage Examples

---

---

Problem and Motivation

In modern medical practice, patient information is often fragmented: test results are stored by one specialist, medical history by another, and daily observations and important details remain only with the patient. This leads to loss of context, complicates decision-making, and hinders building a holistic picture of health.

Moreover, many patients face situations where doctors respond superficially or do not have time to delve into details. There are minor but worrying questions that do not require an urgent visit to the doctor but are important for peace of mind and a conscious approach to health. PersonalDocAI addresses these issues by uniting all user medical data into a single knowledge base and providing personalized consultations considering individual context.

Limitations

Warning: PersonalDocAI is not a medical professional and does not replace consultation with a doctor. All recommendations are for informational purposes only. Always consult a qualified specialist before making health-related decisions.

Caution: The application is not intended for emergency situations. In cases requiring urgent medical attention, immediately contact a doctor or call emergency services.

Note: All data is processed and stored only locally on your device. Make sure you do not lose access to your data and create backups if necessary.

Product Overview and Key Features

Main Features:

• Unified Knowledge Base: Automatic integration and processing of medical documents, test results, questionnaires, and daily user observations.
• RAG System: Search and answer generation considering the context of all available data.
• Personalized Consultations: Recommendations and answers are formed based on the user's individual history and current state.
• Privacy: All data is processed and stored locally, with no transfer to third parties.
• User-Friendly Web Interface: Intuitive chat for communicating with the assistant and managing your data.
• Persistent Memory: Saving chat history and context for consistent support.

Advantages:

• Saves time and reduces the need for frequent doctor visits for minor issues.
• Increases awareness and control over your health.
• Personalized approach and adaptation of recommendations to the specific user.
• Easy integration of new medical data.

Target Audience:
PersonalDocAI is intended for users who want to keep their medical information under control, receive quick and well-founded answers to health questions, and value the privacy of their data.

Architecture and Technical Details

PrivateDocAI uses a multi-layered architecture that separates the user interface, application logic, AI processing, and data storage layers.

Components and Their Interactions

The main system components include:

• StreamlitApp (app.py): User interface built with Streamlit.
• TherapistAgent (agent.py): Core logic of the therapeutic agent.
• RAGSystem (rag.py): Context extraction and processing system.
• PromptBuilder (prompting/prompt_builder.py): Responsible for building contextual prompts.
• memory/: Stores chat history in JSON format.
• data/input/: Directory for input medical data in Markdown format.

Data Flow and Processing Pipeline

The system processes user requests through a multi-stage pipeline:

Technologies Used

The system is built on a modern Python AI/ML stack:

• Runtime: Python ~3.12.
• Web Framework: Streamlit ≥1.47.0 for the user interface.
• AI Orchestration: LangChain ≥0.2.1 for managing LLM workflows.
• OpenAI Integration: LangChain-OpenAI ≥0.1.7 for GPT-4.1-mini.
• Vector Search: FAISS-CPU ≥1.8.0 for efficient similarity search.
• Vector Embeddings: Model all-MiniLM-L6-v2 for creating document and query embeddings.

Asynchronous Architecture

The system uses asynchronous patterns (async/await) to ensure responsiveness and efficient resource usage.

• TherapistAgent: Asynchronous methods for LLM calls.
• RAGSystem: Asynchronous document retrieval.
• memory/: Asynchronous file I/O operations.
• Integration with Streamlit: Asynchronous callback handling.

Step-by-Step Installation and Launch Guide

Requirements

To install and run PrivateDocAI, you will need:

• Python version 3.12 or higher.
• The uv package manager.
• OpenAI API key.

Installation

1. Clone the repository:

   git clone https://github.com/Zegor88/PersonalDocAI.git
   cd PersonalDocAI

2. Create and activate a virtual environment:

   uv venv
   source .venv/bin/activate  # For Unix/macOS
   # OR
   .venv\Scripts\activate     # For Windows

3. Install dependencies:
   It is recommended to install in development mode:

   uv pip install -e ".[dev]"

   Or from requirements.txt:

   uv pip install -r requirements.txt

4. Set up the OpenAI API key:
   You can set the key via an environment variable in the terminal:

   export OPENAI_API_KEY='your-api-key'

   Or by creating a .env file in the project root:

   OPENAI_API_KEY=your-api-key

5. Create necessary directories:

   mkdir -p src/familydoc_ai/memory

Running the Application

1. Prepare data:
   Place your .md files with medical data in the data/input/ directory.
2. Run the application:

   python run.py

   After launch, the application will be available at http://localhost:8501.

Notes:
The run.py file is the entry point for launching the Streamlit application; it loads environment variables and sets the source code path before running app.py. Project dependencies are listed in pyproject.toml.