Welcome to the GitHub repository for the DigH@cktion Hackathon project!
This project is a chat/voice assistant designed to calculate medical scores.
Our solution uses advanced AI technologies like LangChain, LLM RAG with Vector Search, data chunking, embeddings and Chroma vector database to provide contextual responses by processing information from PDFs and URLs. It also supports function calling for automated medical score calculations.
For speech-to-text, it uses the model Whisper from OpenAI, allowing voice interactions with the assistant.
For natural language processing, the project can leverage either an open-source model with Ollama, such as MedLlama2, or a GPT model with an OpenAI API Key. This flexibility allows the project to run locally or connect to the cloud with OpenAI services.
Demo.mp4
DigH@cktion is the 4th edition of the innovation program dedicated to diseases and cancers of the digestive system.
This program is supported by 20 scientific societies, national bodies, working groups, and associations in the ecosystem of digestive system diseases and cancers. The goal of DigH@cktion is to encourage the creation of innovative digital solutions that benefit patients and healthcare professionals.
We are proud to announce that our project won two prizes at the DigH@cktion Hackathon:
- Prix du Public: Recognizing the project as the best solution by public vote.
- Prix Wilco: Awarded by WILCO, indicating high startup potential and innovation.
These awards demonstrate our commitment to creating an impactful and innovative solution for the medical field.
Our project involves a chat/voice assistant designed to aid healthcare professionals in medical score calculations. Here are some of the key technical components and their functions:
- LangChain: Enables structured conversational flows and natural language understanding, allowing the assistant to interact with users in a human-like manner.
- Chroma Database: Acts as the vector database to store and retrieve embeddings from various documents, supporting semantic searches across large datasets.
- LLM RAG: Stands for "Language Learning Model - Retrieval Augmented Generation," providing context by embedding information from PDFs and URLs, allowing the assistant to offer more accurate and informed responses.
- Function Calling: Facilitates the computation of medical scores, allowing the assistant to invoke specific functions based on user requests.
- Whisper: An OpenAI speech-to-text model for voice-based interactions.
- Ollama/MedLlama2 or OpenAI GPT: Supports both local processing with open-source models and cloud processing with OpenAI's GPT model.
To set up the project on your local environment, follow these steps:
- Clone the Repository: Download the source code to your local environment.
git clone https://github.com/AndreIglesias/DigHacktion
- Install the dependencies with poetry.
poetry install
- Activate the Virtual Environment.
poetry shell
- Optionally Create an .env file for the
OPENAI_API_KEY.
This project is licensed under the MIT License.