Image questions
Fine-tuned BLIP predictions through the /predict/ endpoint.
NeuroVision BHPC VQA
Medical-image questions, model training and a research-aware AI assistant.
Overview · Implementation · Evidence
01THE MOTIVATION
The problem
Visual question answering requires reasoning about both medical imagery and a question. This project explores that interaction using research datasets and reproducible experiments; it is not clinical software.
STATUS / Research prototype
02UNDER THE HOOD
The implementation
THE BUILDFine-tuned BLIP on VQA-RAD, served image-and-question predictions through FastAPI, and added a LangGraph agent with PubMed and web-search tools. DVC and MLflow support reproducible training and experiment tracking; a Streamlit interface makes the pipeline accessible.
Research assistant
LangGraph tool calling with PubMed and SerpAPI search.
Reproducible experiments
Dataset processing, training and evaluation tracked with DVC and MLflow.
03ACTUAL ARTIFACTS
Inside the build
Screens from the application, with the workflow and context for each view.
04THE EVIDENCE
Read the source
Explore the code, run the application where available, or watch the original demonstration.