Bhramastra at #SMM4H-HeaRD 2026: A Multi-Stage Hunter-Judge Pipeline using DSPy-Optimized LLMs for Multilingual ADE Detection
Year: 2026
Type: conference-paper
Abstract: This paper describes the submission by Team Bhramastra for the #SMM4H-HeaRD 2026 Shared Task 1, focused on personal Adverse Drug Event (ADE) detection in multilingual social media.We propose Hunter-Judge, a decoupled architecture designed to handle extreme class imbalance and linguistic variance across seven languages, including a surprise zero-shot Farsi set.Our system employs a fine-tuned multilingual mDeBERTa-v3 model as a high-recall filter ("Hunter"), followed by a Gemini-2.5-Flashmodel ("Judge") optimized via the DSPy framework for precisionoriented agentic adjudication.By implementing a reasoning protocol grounded in clinical RAG evidence and universal ingredient mapping, our pipeline achieved the highest average F 1 -score (0.6653) among all teams.Notably, it demonstrated strong zero-shot generalizability on Farsi (F 1 : 0.5863), highlighting the effectiveness of medically-grounded adjudication in low-resource contexts.
Author: Bhaarat Pachori
Language: English
Subfield: Toxicology
Domain: Life Sciences
Open Access status: hybrid
OpenAlex ID: W7166860576