Pediatric Emergency Medicine Research Agent

This project implements a command-line research assistant for answering complex clinical questions using retrieval-augmented generation (RAG). The system performs dense retrieval over a domain-specific medical corpus and generates answers using a large language model (LLM).

Project Purpose

This tool is designed to simulate a research assistant that can:

• Retrieve relevant scientific and clinical literature
• Reason over the retrieved documents to answer multi-hop questions
• Support clinicians or medical researchers in pediatric emergency medicine

The project showcases modular, domain-adapted RAG pipelines with strong retrieval and synthesis capabilities, intended for demonstration, research, or extension.

Technology Stack

• Retrieval: FAISS + Sentence Transformers (e.g., gte-multilingual-base)
• LLM Reasoning: OpenAI gpt-4o or gpt-3.5-turbo
• Corpus: Scientific and clinical documents (e.g., from Hugging Face scientific_papers)
• Interface: Command-line application

File Descriptions

File/Directory	Description
main.py	Entry point that loads the corpus, builds the FAISS index, and runs the query interface.
agent/embed.py	Encodes documents into dense vector embeddings using a sentence transformer.
agent/load_corpus.py	Loads and optionally preprocesses the corpus for indexing.
agent/search.py	Performs dense vector similarity search using FAISS.
agent/reasoner.py	Sends the user query and retrieved documents to the OpenAI API for LLM-based reasoning.
corpus/	(Optional) Folder for storing local or custom text corpora.

Example Usage

$ python main.py
Enter your research question: 
A 3-week-old infant presents with vomiting, hypotension, hyponatremia, hyperkalemia, and hypoglycemia. What is the likely diagnosis and initial management?

Agent Answer:
The presentation is consistent with congenital adrenal hyperplasia due to 21-hydroxylase deficiency...

Setup Instructions

1. Clone the repository and create a virtual environment:

git clone https://github.com/your-username/research-agent.git
cd research-agent
conda create -n research_agent_env python=3.9
conda activate research_agent_env
pip install -r requirements.txt

2. Set your OpenAI API key:

export OPENAI_API_KEY=your-key-here

3. Run the agent:

python main.py

Future Improvements

• Support hybrid retrieval (dense + sparse)
• Improve domain coverage with PubMed abstracts
• Add chunking for long documents
• Optional web UI using Streamlit or FastAPI

Relevance

This project demonstrates practical experience with:

• Retrieval-augmented generation (RAG) systems
• Dense vector search with FAISS
• LLM prompt engineering and OpenAI integration
• Scientific and medical domain adaptation

It may be of interest to teams building AI agents for healthcare, question answering, or knowledge retrieval tasks.