Rosie

TLDR

• Paul Conyngham’s dog had months to live. He had no biology degree.

• He used ChatGPT, AlphaFold, and Grok to design a personalized mRNA cancer vaccine. The tumor shrank 75% in a month.

• Anyone who has sat in a hospital waiting room, scanning the same web pages, reading the same research abstracts that lead nowhere, knows the particular helplessness of having everything to give and no way to give it.

• That helplessness now has a crack in it.

• Tenacity and the right questions will now take you further than any of us thought possible. Your own genomic data, your own scan, your own case: all of it is now a puzzle you can actively work.

• This is the first real biohack. And the first of many.

---

In December 2025, Paul Conyngham drove ten hours to a veterinary clinic outside Brisbane. His rescue dog Rosie, an eight-year-old Shar Pei, had been given one to six months to live. Aggressive mast cell cancer. Multiple tumors. Conventional chemo had slowed the spread and done nothing more. The vets had run their course.

Conyngham was a data engineer. Seventeen years in machine learning and AI consulting. Zero background in biology. When the diagnosis came, his response was blunt: “No worries. I’m a data analyst. I’ll figure this out with ChatGPT.”

Six weeks after Rosie’s first injection, she was at the dog park. She spotted a rabbit. She jumped the fence.

The Waiting Room

Everyone reading this knows someone who has had cancer. One in three people will face it themselves at some point in their lives.

And most of us know the particular feeling that arrives shortly after the diagnosis. You sit with someone you love and you want to do something. So you start searching. You read the same clinical summaries. You find forums where other families are doing the same thing. You print articles and bring them to appointments. Doctors are kind and patient and the answers are always some version of the same thing: this is what we have, this is what we can do, here are the odds.

Tenacity counted for very little in that room. Love counted for very little. The knowledge was gated, the tools were out of reach, and the most you could do was advocate loudly within a system that moved at its own speed.

Conyngham walked into that same wall. And then he walked around it.

What He Did

The process had three phases, each with a different AI model doing a different job.

He started with ChatGPT as a research navigator. It pointed him toward immunotherapy and identified the Ramaciotti Centre for Genomics at the University of New South Wales as the place to sequence Rosie’s DNA. He paid $3,000. The Centre compared her healthy cells to her tumor cells and mapped exactly where the mutations had taken hold. The output was gigabytes of raw genomic data.

Then came AlphaFold, Google DeepMind’s protein structure prediction model, the one that won the Nobel Prize in Chemistry in 2024. Proteins execute virtually everything that happens inside a cell. Cancer hijacks them. AlphaFold takes a gene sequence and predicts the precise three-dimensional shape of the protein it encodes, a task that used to require years of specialized lab work. Conyngham used it to model c-KIT, the mutated protein driving Rosie’s cancer. He identified which mutations were most likely to trigger an immune response. He ranked them.

The final vaccine construct was designed by Grok. Codon-optimized. Stability features built in. The output was half a page of formulas.

Conyngham handed that half page to the UNSW RNA Institute. They synthesized the vaccine in under two months.

Within a month of the first injection, the tennis-ball-sized tumor on Rosie’s leg had shrunk by 75%.

Martin Smith, director of the Ramaciotti Centre, recalled his reaction when Conyngham first called: “Paul was relentless. He told me he’d analyzed the data, found mutations of interest, used AlphaFold to find the mutated proteins, identified potential targets, and matched them to drugs. I’m like, woah. That’s crazy.”

A Few Things Worth Naming

Rosie still has cancer. UNSW has been clear: the disease is incurable. One tumor responded dramatically. Another did not, and Conyngham is already designing a second vaccine targeting its specific mutations.

The treatment required a checkpoint inhibitor administered alongside the vaccine. The mRNA construct alone was not the complete picture.

The $3,000 covers the sequencing. The researchers at UNSW and the University of Queensland contributed time and expertise that would price this experiment far higher on the open market.

The First Real Biohack

Biohacking has been around as a term for years. It meant something modest: tracking your glucose, optimizing your sleep, experimenting with supplements. A subculture of curious people trying to improve their own biology, one measurement at a time.

Conyngham did something categorically different. He took his dog’s raw genomic data and directed AI systems to interrogate it, model the proteins it encoded, identify the mutations worth targeting, and design a molecular blueprint for a treatment. He orchestrated a research pipeline that would have required a fully equipped lab, a team of specialists, and years of institutional access.

He ran it from his computer in months.

The science was not new. mRNA technology has been in development for decades. The COVID vaccines brought it into every household’s vocabulary. Moderna, Merck, and BioNTech have personalized cancer vaccine programs running in late-stage clinical trials right now. The tools existed. AlphaFold has been publicly available. Grok is accessible to anyone with an account.

What was new was someone with the right mental model for directing AI systems, applying that skill to a biological problem, and producing a result.

The credential that used to be required was a decade of specialized training and institutional access. The credential now is knowing how to ask.

Your Genome Is Now Readable

Your medical data has always technically belonged to you. Your scans, your blood work, your genomic sequence, your biopsy results. You could carry them from one appointment to another. They sat in folders and hospital systems and lab databases.

Practically speaking, they were a language with no translator.

Conyngham received gigabytes of raw sequencing data and turned them into a research instrument. He asked questions of his own dog’s genome that no oncologist had the bandwidth to ask for a single patient. He ran hypotheses in hours that a research institution would have queued for months.

That relationship between a person and their own biological data is new.

AI can already interrogate a scan or an X-ray in ways that would have required a specialist consultation two years ago. Point it at your own imaging and ask it questions. Ask about the density patterns your radiologist mentioned in passing. Ask what the literature says about a finding in someone your age. You will not always get a definitive answer. You will get further than you could before.

And that distance is growing.

For anyone who has ever sat in that waiting room with a folder full of printouts and nowhere to put their drive, this matters.

What This Cracks Open

The global oncology market is worth roughly $250 billion today, projected to exceed $600 billion by 2034. Cancer medicine alone accounts for nearly 20% of all global pharmaceutical spending.

That industry is built on a structural assumption: that custom treatment is economically impossible. A drug requires years of trials, regulatory approval, and a patient population large enough to justify the investment. The individual patient has always been a rounding error in a much larger calculation. You get what works for most people. You bend your body around what’s available.

Chemotherapy is the clearest expression of that logic. It kills fast-dividing cells, cancerous and healthy alike, because distinguishing them precisely enough was beyond what the tools allowed. An industrial-era solution. Effective enough to justify the damage it causes.

Rosie’s vaccine was designed for her DNA. Her mutations. Her neoantigens. For her body and no other.

We wrote about this moment when it arrived in software. When we described one of one at scale, software becoming bespoke and custom-built for a single company’s exact operations, we used biotech as the analogy. In 2001, sequencing a human genome cost $95 million. Today it costs roughly $200. When the cost of a fundamental input collapses, the economics of everything built on top of it change. We used it as an analogy. It was a preview.

The research phase of pharmaceutical development, target identification, protein modeling, construct design: that has always been the most protected, most credentialed part of the pipeline. Conyngham ran it from his computer, using publicly available tools, in months. The manufacturing still required a lab. The administration still required ethics approval. The intellectual work moved.

Pharmaceutical labs are extraordinary at what they do. Precision synthesis, quality control, supply chain infrastructure at scale. What is changing is who arrives at the lab door with the design.

The Last Wall

Conyngham spent three months writing a hundred-page ethics application, two hours every evening, before Rosie could receive a single injection.

He said the regulatory process was harder than the science.

The frameworks governing pharmaceutical development were built for a world where research took years, required institutional infrastructure, and produced treatments for mass deployment. A one-patient, AI-designed, individually synthesized treatment fits none of those assumptions. There is no existing pathway for it.

The institutions that build evaluation frameworks for individualized treatments will shape where this science gets practiced. The gap between what is technically possible and what is institutionally permissible is already wide. It will only get wider.

The First Breach

The coverage of this story has been warm and wide. Man loves dog. Man refuses to accept the odds. Dog jumps a fence. It is a moving story.

It is also a precedent. The first documented instance of an individual, outside institutional research, using publicly available AI tools to design a personalized treatment for a specific patient’s specific mutations, having it synthesized, administering it through proper channels, and producing a measurable clinical result.

Every step of that chain is now documented. The methodology. The tools. The regulatory pathway. The cost. The timeline. All of it is public, and all of it will be studied by everyone who comes next.

Conyngham is already working on a second vaccine for Rosie’s unresponsive tumor. He is talking to the teams involved about how to help others access similar approaches. The UNSW RNA Institute director said the approach could extend to neurological diseases. Martin Smith asked the question out loud: if we can do this for a dog, why aren’t we doing it for all humans with cancer?

The first time something happens, it is extraordinary.

The second time, it is a data point.

The tenth time, it is a methodology.

We are at the first time.

Watch the velocity from here.

Rosie is chasing rabbits.

---

We are Exponential Partners. We guide leadership teams through AI transformation, from understanding what’s possible to building what matters. If you are exploring what AI means for your organization, every engagement starts with a 30-minute conversation. Reach us at hello@exponentialpartners.io.