Pride-Boiler at MedGenVidQA 2026: LLM-Augmented BM25 Retrieval with Corrective Self-Verification for Biomedical Evidence Retrieval

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Health & Medical Research · Depth: Expert, medium

Summary

The Pride-Boiler system, submitted to MedGenVidQA 2026 Shared Task A, addresses the challenge of retrieving relevant biomedical articles and instructional videos for consumer health queries. This system integrates Pyserini BM25 retrieval with Google Gemini-powered query rewriting and a corrective self-verification loop, inspired by the CRAG paradigm. It generates clinically optimized search texts for PubMed abstracts (using MeSH terms) and video subtitles (using procedural language). After BM25 retrieves candidates, Gemini scores them, blending its judgment with lexical signals. A quality grader assesses top results, triggering up to three corrective cycles with reformulated terminology if deemed insufficient. Orchestrated as a LangGraph state machine, Pride-Boiler achieved first place in PubMed article retrieval, with an nDCG of 0.6532 and MAP of 0.5550, surpassing the Text-RR baseline. For video retrieval, it scored 0.5304 in MAP and 0.5927 in nDCG, outperforming other systems but falling below the baseline due to noisy subtitle text. The pipeline code is available on GitHub.

Key takeaway

For Machine Learning Engineers developing biomedical information retrieval systems, this work demonstrates a robust approach. You should consider integrating LLM-driven query optimization and a corrective self-verification loop, similar to the CRAG paradigm, into your BM25 pipelines. This strategy significantly boosts PubMed article retrieval performance, as shown by Pride-Boiler's top ranking. However, be aware that lexical matching limitations persist for noisy data like video subtitles, requiring alternative strategies for such content.

Key insights

LLM-augmented BM25 with corrective self-verification significantly improves biomedical evidence retrieval from PubMed.

Principles

• LLMs can optimize queries for specific data types.
• Self-verification loops enhance retrieval quality.
• Blending lexical and LLM signals improves relevance.

Method

The system uses Google Gemini for query rewriting and candidate scoring, Pyserini BM25 for initial retrieval, and a LangGraph-orchestrated corrective loop for refinement.

In practice

• Use Gemini to generate MeSH terms for PubMed queries.
• Implement a CRAG-inspired loop for result validation.
• Combine LLM relevance scores with BM25 lexical signals.

Topics

• Biomedical Information Retrieval
• LLM-Augmented Retrieval
• BM25
• Corrective RAG
• Google Gemini
• MedGenVidQA 2026

Code references

• basilll007/BioNLP

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• Benchmarking Retrieval Strategies for Biomedical Retrieval-Augmented Generation: A Controlled Empirical Study — Cross-Encoder Reranking significantly improves biomedical RAG performance by enhancing contextual precision.
• The State of Information Retrieval in 2026 — Information retrieval has evolved into a reasoning problem, integrating LLMs and agentic search for complex, context-aware query processing.
• An LLM as arbiter in RAG retrieval: picking the right candidate with reasons — An LLM arbiter, given structured candidate briefs, can rank RAG results with auditable reasons, surpassing score fusion.
• Retrieval-Augmented Generation for Clinical Question Answering in Portuguese Drug Leaflets: Benefits and Limitations — RAG improves factual recall in clinical Q&A but needs adaptive control for complex reasoning to avoid anchoring bias.
• T2D-Bench: Evidence-Gated Evaluation of LLM Outputs for Type 2 Diabetes Using a Multi-Layer Clinical-Lifestyle Knowledge Graph — LLMs require evidence-gated evaluation against clinical knowledge graphs to ensure medical accuracy and guideline compliance.
• TrialMatchAI: an end-to-end AI-powered clinical trial recommendation system to streamline patient-to-trial matching — TrialMatchAI automates clinical trial matching using LLMs and RAG for explainable, accurate patient eligibility.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.