SePA Web Search Pipeline

Implementation of the web search component from "SePA: A Search-enhanced Predictive Agent for Personalized Health Coaching" (IEEE BHI 2025).

Repository Structure

paper-version/ - Original implementation from the paper

• Single-query search, CPU-based processing
• Use for reproducing paper results

improved-search/ - Improved implementation

• Multi-query expansion, GPU acceleration, RRF ranking, improved fetching and processing of sources.
• Use for production deployments

Key Enhancements

1. Multi-Query Expansion: Generates 4 diverse queries (general, scientific, demographic-specific, practical)
2. Reciprocal Rank Fusion: Merges results with k=60, 2x boost for trusted medical/sports domains
3. GPU Acceleration: 10x faster embeddings via HuggingFace Inference Endpoints
4. Advanced Content Fetching: Headless browser support for JavaScript-rendered content

Citation

@article{ozolcer2025sepa,
  title={SePA: A Search-enhanced Predictive Agent for Personalized Health Coaching},
  author={Ozolcer, Melik and Bae, Sang Won},
  journal={arXiv preprint arXiv:2509.04752},
  year={2025}
}

Notes

• Paper version preserves the exact implementation for reproducibility
• Improved version includes production optimizations not described in the paper
• See individual directories for detailed documentation