Context
Pathologists sometimes get a whole slide image (WSI) they can’t diagnose with confidence. Aignostics built PathoSearch for those cases: a reverse image search for pathology. The pathologist uploads a screenshot of a WSI and gets back similar cases with their diagnoses.
This was the company’s first SaaS product. I joined a cross-functional team of 6 to build the frontend in a stack new to the company: Next.js and Supabase.
Solution
Live search results. The model took time to process each image. I used Supabase real-time subscriptions so results appeared in the browser as soon as they hit the database, with no polling and no refresh.
Slide viewer. With tight deadlines, building a WSI viewer from scratch was out of the question. I integrated the open-source DICOM Microscopy Viewer and added auto-zoom to the regions of interest where a similar image matched.
Pipeline repair. A major refactor had broken the CI/CD pipeline and the Playwright E2E tests, and they stayed broken for a while. I fixed the tests and got the pipeline fully green again.
AI-assisted workflow. I trialed GitHub Copilot Agents on minor tickets: the agent opened PRs and developers reviewed them. This took simple tasks off developers and let the product owner request small changes directly.
The MVP received positive feedback from management and external stakeholders, and was approved for full development.
Reflections
This was my first time using Supabase. The developer experience is excellent, and Supabase Studio made inspecting data and iterating on the schema fast.