Gamgee Turns a Viral ChatGPT Dog-Cancer Story Into a YC Startup

What actually happened

Paul Conyngham is a machine learning engineer with 17 years in data science, no formal biology training, and a rescue dog named Rosie whose mast cell tumors had stopped responding to surgery, chemotherapy, and immunotherapy. Over several months he analyzed the genomic differences between Rosie's healthy and tumor cells — comparing them to find cancer-specific mutations — using ChatGPT to work through the analysis and AlphaFold to help model the resulting protein structures, then collaborated with researchers at the University of New South Wales and the University of Queensland to turn that into a bespoke mRNA vaccine. Rosie was treated over the 2025 Christmas break; within a month, her tennis-ball-sized tumor had reportedly shrunk 75%.

From one dog to a company

Conyngham has now launched Gamgee, backed by Y Combinator's Summer 2026 batch, to offer personalized mRNA cancer vaccines for other dogs. The pitch: sequence a dog's tumor and healthy DNA through veterinary partners, identify the mutations unique to that cancer, design a matched vaccine, and coordinate manufacturing and delivery through a treating oncologist. Gamgee's stated goals are a 48-hour design turnaround and an end-to-end cost near $2,000 — Conyngham says he spent about $3,000 on sequencing alone for Rosie's case. The company is currently running a 30-dog clinical trial in Australia with veterinary oncologists, and says it's accepting cases globally. It's a two-person team so far, aimed at a real market: roughly six million dogs are diagnosed with cancer in the US every year.

The caveats that matter

One dog is an anecdote, not a study, and independent scrutiny has said so plainly. Writing in The Conversation, biomedical sciences professor Justin Stebbing noted that Rosie's case has no control group, that mast cell tumors are known to behave unpredictably on their own, and that there's no peer-reviewed data confirming how much of her improvement is attributable to the vaccine, how durable it is, or whether the approach would work for a different dog. He was also careful to point out that the AI didn't design a working vaccine unsupervised — qualified scientists at two universities checked and validated its output at every step. That's exactly what Gamgee's 30-dog trial exists to test at scale.

Why this is a vibecoding story, not just a pet story

Strip away the dog and what's left is a familiar shape to anyone building with AI tools: a domain expert wasn't the bottleneck — the pipeline was. ChatGPT handled orchestration and analysis across a field Conyngham had never worked in, AlphaFold supplied the specialist structural prediction, and human experts sat at the end of the chain to validate before anything touched a living patient. That division of labor — AI compresses the time-to-first-working-pipeline, humans own the final sign-off — is the same discipline that matters anywhere an LLM sits in a consequential loop, trading systems included. The lesson isn't that a chatbot designed a cancer vaccine. It's that the distance between "non-specialist with a good AI workflow" and "credible technical prototype" keeps shrinking, and the review step in between is doing all the safety work.

What to watch next

• Whether Gamgee's 30-dog Australian trial produces outcomes that hold up statistically, not just anecdotally.
• Whether any of this work reaches peer-reviewed publication — none has been published as of this writing.
• Whether Gamgee's $2,000 / 48-hour targets survive contact with running the pipeline at case volume rather than for one dog.