Researcher identification (required for PubMed NCBI compliance)

RESEARCHER_EMAIL=your.email@university.edu

Integration tests (requires network)

python -m pytest tests/ -m integration -v

PR Requirements

• All existing tests must pass
• New features need ≥85% coverage
• Type hints on all new functions
• Google-style docstring on all new public functions
• Update CHANGELOG.md entry

Installation

Via pip (recommended): ```bash

Core install

pip install eeg-rag

Setup Dev Environment

pip install -r requirements-dev.txt

Example output:

🚀 Quick Start

💻 Usage

Quick start: ~1,000 papers (5–10 min)

python scripts/run_ingestion.py --sources pubmed arxiv

Configuration

```bash

Copy environment template

cp .env.example .env

Edit `.env`:

Optional: speeds up ingestion significantly

NCBI_API_KEY=your_ncbi_key_here # https://www.ncbi.nlm.nih.gov/account/settings/ S2_API_KEY=your_s2_key_here # https://www.semanticscholar.org/product/api

LLM for synthesis (optional — system works without it for retrieval)

OPENAI_API_KEY=sk-your-key-here OPENAI_MODEL=gpt-3.5-turbo

Optional services

NEO4J_URI=bolt://localhost:7687 NEO4J_USER=neo4j NEO4J_PASSWORD=your_password REDIS_HOST=localhost REDIS_PORT=6379 ```

With REST API server (FastAPI + uvicorn + SSE)

pip install "eeg-rag[api]"

Start the API Server

```bash

API Endpoints

[!NOTE] Interactive docs available at http://localhost:8080/docs once the server is running. No API key required for retrieval-only queries.

<details> <summary>📡 Full curl Examples — Click to Expand</summary>

Standard search:

curl -X POST "http://localhost:8080/search" \
  -H "Content-Type: application/json" \
  -d '{"query": "deep learning EEG seizure detection", "max_results": 10, "synthesize": true}'

Streaming search (SSE):

curl -N -X POST "http://localhost:8080/search/stream" \
  -H "Content-Type: application/json" \
  -d '{"query": "P300 amplitude in Alzheimer disease", "max_results": 5}'

Paper details:

curl -X POST "http://localhost:8080/paper/details" \
  -H "Content-Type: application/json" \
  -d '{"pmid": "28215566"}'

Health check:

curl http://localhost:8080/health

</details>

<p align="right">(<a href="#top">back to top ↑</a>)</p>

---

Python SDK

```python import requests

ERP Components

Multi-Stage Pipeline

The RRF score formula:

$$\text{RRF}(d) = \sum_{r \in R} \frac{1}{k + r(d)}$$

where $k=60$ (default), $r(d)$ is the rank of document $d$ in ranker $r$, and $R$ is the set of retrieval methods. This provably outperforms linear score combination (Cormack et al., 2009).

Retrieval Stage Comparison

PubMedBERT vs Alternatives

</details>

<p align="right">(<a href="#top">back to top ↑</a>)</p>

---