Docs & Writing
Pipeline explainers, architecture notes, and the Gemma4Good hackathon writeup.
The cancer vaccine story
How one ML scientist spent three months designing a personalized cancer vaccine for his dying dog — and what it revealed about the gap between research and accessibility.
Project Rosie: Open Source AI Pipeline
Gemma4Good hackathon writeup. How Gemma 4's native function calling, multimodal vision, and 256K context make it the intelligence layer — not a report formatter bolted on at the end.
From DNA to vaccine candidates
The full pipeline journey, written for readers with no biology background. Think of it as tracing a request through a distributed system — except the data is a dog's DNA.
Key architecture decisions
Every non-obvious choice made in Project Rosie, and why. Written so a future contributor or hackathon judge can understand the reasoning without asking.
Frontend architecture
The Next.js 16 + Supabase + Tailwind v4 stack explained. Written for engineers unfamiliar with the frontend. Biology background not needed.
Cloud deployment architecture
How the pipeline runs in the cloud: GCS uploads, Cloud Run Jobs, Workload Identity Federation, and the callback pattern that drives live status updates in the browser.
Canine data end-to-end run
How the first real canine case ran through the pipeline: VEP v115 annotation with ROS_Cfam_1.0, pVACseq with DLA alleles, the Docker WSL2 volume mount workaround, and the PIK3CA V125M neoantigen that came out the other end.
Source code
Full pipeline, frontend, and scripts on GitHub.