Context
Aignostics builds AI-powered pathology tools for drug discovery and diagnostics.
Pathologists view digitized tissue scans, called Whole Slide Images (WSIs), and AI analysis results in the browser to look for cancerous cells. We had several viewers for this including open source and in-house developed solutions.
These viewers had different backend APIs, features, and styling. Each needed its own maintenance, and they couldn’t scale to displaying millions of AI-generated polygons, as required by new products.
Therefore, I led the effort to unite these viewers under one solution that would scale, together with one other frontend developer.
Solution
First, we turned the product requirements into decision criteria, reviewing them with frontend, backend, product, and management. We prototyped four options: extend the in-house viewer, adopt open source, buy, or build a new viewer.
After prototyping, we wrote and doc-reviewed an Architecture Decision Record (ADR) capturing the tradeoffs. We proposed building a new viewer based on deck.gl, supporting GPU-accelerated rendering of millions of polygons. Using its composable layers and widgets allowed each project to import only what it needs, apply its own styling, and support loading data from different backend APIs.
To set a high quality bar from the start, we required and implemented over 90% test coverage. I wrote a Claude Code AGENTS playbook. GitHub Copilot Agents handled code reviews and OX Security handled security reviews.
We delivered the first WSI viewer as a composable React library that could load WSIs via DICOMWeb and AI analysis results via PMTiles, rendering 3 million polygons at 60 fps and cutting maintenance down by 67%.
Reflections
This being a high-priority project, early transparency and doc review meetings paid off big time. Stakeholders and management stayed aligned, agreed on evaluation criteria up front, and raised issues early.
Late in the ADR review, we were still asked to show that our proposal would meet a hard deadline. We stepped back and made rough estimates. They showed we would make the deadline and deliver more long-term value. Indeed, we shipped on time.