Systematic literature reviews are essential for evidence-based research, but they are often slow and require a lot of effort. Motivated by the potential of Retrieval-Augmented Generation (RAG) technology, our team from Georgia Tech's Master in Computer Science program, with early guidance from researchers at Robert Gordon University, created PICO Scholar. This AI-powered platform helps researchers by turning their queries into embeddings, matching them with a semantic vector database, and pulling out important research details—Population, Intervention, Comparison, Outcome (PICO). It ranks relevant studies and offers a chatbot powered by RAG for exploring documents more deeply. Future updates aim to add features for real-time collaboration, making it easier for teams to work together on reviewing literature.

PICO Scholar uses a solid microservices setup with Red Hat OpenShift AI, boosted by Intel’s OpenVINO toolkit. It runs models like SciBERT for embeddings and TinyLlama for generating text, managed by ModelMesh. LlamaIndex organizes embeddings into a TiDB VectorStore for smarter searches, and fine-tuned adapters improve accuracy for specific fields. MinIO S3 supports smooth model storage, and the platform’s microservice design makes it scalable and resource-efficient.

What makes PICO Scholar stand out is not just its technology but its focus on community and collaboration. Built during two hackathons - including a 2nd place win at the Red Hat & Intel AI Hackathon - with changing team members and diverse skills, the project follows open-source principles, encouraging contributions from all backgrounds. Our goal is to make powerful research tools accessible to everyone, helping researchers worldwide speed up discoveries. By continuing to improve PICO Scholar, we hope to support the scientific community and build a platform that grows and evolves to meet its needs.

PICO Scholar GitHub repository