Projects
GAA Simulation Platform
An adversarial training simulation platform where AI agents play patients and examiners, so trainees can practice difficult clinical conversations and be assessed consistently. Scenarios, agents, and scoring methods are developed in the open.
OSCE-Project
Medical dialogue evaluation framework inspired by Objective Structured Clinical Examinations. Simulates 64 patient personas measuring Empathy, Persuasion, and Safety.
FHIR-Agent
An LLM agents framework for interacting with FHIR databases, enabling structured clinical data retrieval and reasoning.
OSCE-AgentBeats Leaderboard
Public leaderboard for the OSCE evaluator within the AgentBeats challenge ecosystem. Compare medical agent performance across standardized clinical scenarios.
AgentBeats Security Arena
Adversarial security testing framework for AI agents. Evaluates robustness and safety against prompt injection and manipulation attacks.
OSCE Real-Time Voice
Real-time voice-based clinical examination system extending the OSCE framework with speech interaction for realistic doctor-patient dialogue evaluation.
Model Training & Distillation
Fine-tuning and knowledge distillation pipelines for domain-specific medical language models, aiming for smaller and faster models that run on local hardware.