Collaborative project, contributed key production

Collaborative project, contributed key production components of Med-V²QA, a clinical Visual Question Answering system built on the MUMC architecture for radiology image analysis.

My contributions: built the Dual-Gate Guardrail System (Gate 1 rejects non-medical intent, Gate 2 uses a local CLIP classifier to reject non-medical images ensuring clinical safety and system focus); developed the Batch Triage module using zero-shot CLIP to automatically sort and prioritise the most abnormal scans from uploads of up to 20 images; integrated Voice I/O using OpenAI Whisper for speech-to-text transcription and gTTS for audio readout of answers; built the PDF report generation module producing structured clinical reports with patient data, notes, and VQA results; and developed the FastAPI backend handling high-performance async request processing across all services.

Built as part of a team my work covered the safety layer, multimodal I/O, clinical reporting, and backend API infrastructure.