Projects

A chest X-ray research tool that runs machine-learning inference in the browser, so uploaded images remain on the user's device.

• Uses an InceptionV3 model trained on 100,000 NIH ChestX-ray14 images to score 11 thoracic findings plus No Finding and generate model-attention heatmaps.
• Runs with ONNX Runtime Web and reached a 0.807 macro AUROC on a patient-separated NIH test set of 2,500 images.

A conversational assistant that helps event organizers turn an event idea into participant surveys and personalized human-bingo boards.

• Uses session memory, prompt guardrails, and schema-driven tool calls to return structured survey and board content.
• Integrated the assistant into Blossom's live workflow and compared it with the original event-creation process through user testing.