FastAPI Backend Logic

📌 Project Overview

FASTAPI Backend Server Code to efficiently handle code for Mini-CDSS Project.

📖 Description

Mini-CDSS helps doctors to take an initial patient encounter and creates an initial patient report, give diagnosis and even use Tavily Web Search to find all the relevant Best Practises. The Doctor can also extract medical insights from any uploaded documents before processing with the reporting system. We also allow the doctor to RAG - Chat with the uploaded documents. The Doctor can also use our vision model to upload any images and query from it, give feedback etc.

AIM: Founding Steps of Next-gen CDSS systems, incorporating Gen-AI in report building, multimodal capabilities, CDSS is a highly rigid rule based system, Gen-AI introduces flexibility, context-awareness for fundamental systems to work on, and leveraging frameworks like Langgraph and Pydantic it is possible not only to reduce LLM Hallucination but also introduce dynamic real-time decision branches based on simple text.

🏗️ Tech Stack

• Language: Python 3.x
• Framework: FastAPI
• GenAI Framework: Langgraph
• Generative Services Used: Gemini, Groq
• Web Search Tool Tavily
• Database(Vector-DB): ChromaDB [persisted on hosted server;linked with session uuid for a duration of 15 minutes before the db is flushed]

📂 Project Structure

.
├── config
│   ├── __init__                 # Init Package
│   ├── credentials              # Loads in base-64 encoded Credentails of Google Service Account (needed to run gemini on server.)
│   ├── fastapi_models           # API Endpoint pydantic structure
│   ├── hugging_face_ner         # Runs Hugging Face model on intial patient encounter data to extract relevant entities, and returns an entity dict.
│   ├── main_graph               # The main code dedicated to NER Report, Prelim Report, Feedback Loop and Best Practise Report Generation Logic.
│   ├── medical_summarizer_graph # Extracts the relevant medical details from uploaded documents.
│   ├── rag.py                   # A REACT Blueprint of a smarter RAG Agent.
│   ├── validate_api.py          # Function to validate the entered API Keys.
│   ├── vectordb.py              # Function dedicated to creation and deletion logic of Vector-DBs.
│   ├── vision_graph.py          # Brings Multi-modal capabilities to our project.
│
├── cron                         # Server side Background tasks.
│   ├── __init__                 # Init Package
│   ├── storage                  # Keeps track of vector_db_list, tracking current Vec DBs on the server and their lifetime.
│   ├── jobs                     # Secheduled Jobs that triggers at a given moment/interval.
│   ├── tasks                    # Logic of the Jobs scheduled.
│
│
├── .env                         # Tracks the env variables and their details
├── main                         # Main Server Side logic with endpoints.
├── requirements                 # Necessary dependencies to run the project
├── render.yaml                  # COnfig file to run Server on render.

1️⃣ Clone the Repository

git clone https://github.com/Recker-Dev/Mini-CDSS-FastAPI.git
cd Mini-CDSS-FastAPI

2️⃣ Set Up Virtual Environment

conda create --name test_env python=3.10 -y
conda activate test_env

3️⃣ Install Dependencies

pip install -r requirements.txt

4️⃣ Configure Environment Variables

Create .env file and put

• GOOGLE_CREDENTIALS_BASE64 [Service Account credentials, gives a JSON file, need to encode it as a b64 format and decrypt it to be used normally]
• GOOGLE_API_KEY
• TAVILY_API_KEY
• GROQ_API_KEY
• HF_TOKEN

6️⃣ Start the FastAPI Server

uvicorn app.main:app --reload

7️⃣ Access API Documentation

• Swagger UI: http://127.0.0.1:8000/docs
• ReDoc: http://127.0.0.1:8000/redoc

📜 API Endpoints

Method	Endpoint	Description
POST	/validate_and_set_api	API configuration
POST	/set_api	Helps to set the keys just before graph starts.
POST	/graphstart/	Used to start main graph
POST	/prelimInterruptTrigger	Takes feedback from user based on prelim diagnosis report generated.
POST	/nerReport	Returns the NER Report of patient encounter.
POST	/prelimReport	Returns the Prelim Report of patient encounter.
POST	/bestpracReport	Returns the Best Practises Report of patient encounter.
POST	/addFilesAndCreateVectorDB	Logic to create a vector DB and overwrite if existing.
POST	/ragSearch	Used to search the docs.
POST	/ragAnswer	Returns the rag answer.
POST	/extractMedicalDetails	Used to get Medical Details from the uploaded documents.
POST	/medicalInsightReport	Returns the Medical Insights gathered from the uploaded documents..
POST	/input-image/	Used to input image to vision model.
POST	/input-query/	Enter the query about the image.
POST	/vision-answer/	Returns the answer generated.
POST	/vision-feedback/	Takes feedback for vision output.

📝 License

This project is licensed under the MIT License.

📬 Contact

For any issues or suggestions, reach out at reckerdev@gmail.com.