The Story That Broke the Internet
In early 2026, Paul Conyngham — a machine learning consultant based in Sydney, Australia — did something that most oncology researchers would have called impossible for a non-specialist. He used ChatGPT, AlphaFold, and genomic sequencing data to design a custom mRNA cancer vaccine for his rescue dog Rosie, who had been diagnosed with aggressive mast cell cancer and given only months to live. The tumor on Rosie’s back leg shrank by more than 50% within one month of her first injection. (Qbitai)
This is not just a heartwarming pet story. It is a signal event at the intersection of AI, biotechnology, and personalized medicine — one that forces us to rethink who gets to participate in cutting-edge medical research and how fast breakthroughs can happen when AI tools are freely available.
Who Is Paul Conyngham?
Conyngham runs a machine learning consultancy in Sydney. His day job involves data pipelines and prediction models. He has no formal training in biology, oncology, or any medical field. Rosie, a Staffordshire Bull Terrier and Shar Pei cross, was adopted from a Sydney shelter in 2019 after being abandoned in bushland. When large tumors appeared on her back leg in 2024, the diagnosis was mast cell cancer — the most common skin cancer in dogs and notoriously difficult to treat once it spreads. (Generative AI Publication)
Conyngham spent thousands on veterinary chemotherapy and surgical intervention. The treatments slowed the cancer but did not shrink the tumors. That’s when he started thinking about a different approach — one that leveraged the AI tools he already knew how to use.
The Technical Pipeline: From Sequencing to Injection
What makes this case remarkable is the reproducibility of the workflow. Conyngham didn’t invent new technology. He assembled publicly available tools into a pipeline that, until recently, would have required a multimillion-dollar lab and a team of PhD scientists.
| Step | Tool/Method | Purpose |
|---|---|---|
| Diagnosis confirmation | Veterinary oncology | Confirmed mast cell cancer |
| Genomic sequencing | UNSW Ramaciotti Centre for Genomics | Identify tumor-specific mutations |
| Data processing | Custom algorithms | Process raw sequencing data |
| Protein structure prediction | AlphaFold (DeepMind) | Identify mutant proteins and therapeutic targets |
| Vaccine design | ChatGPT + custom code | Design mRNA sequence targeting neoantigens |
| Synthesis and administration | UNSW RNA Institute | Manufacture and inject the vaccine |
| Regulatory approval | Ethics board (3-month process) | Ensure safety and legal compliance |
The total cost: approximately $3,000 in DNA sequencing plus a ChatGPT subscription. The timeline: weeks instead of the 12 to 18 months typical in traditional biotech pipelines. (Blockchain.news)
Related: How to Use AI Without Getting Fired: A Professional’s Guide (2026)
Professor Sudharsan at the UNSW RNA Institute noted that “a data engineer with no biology background had successfully written the mRNA formula” — and described the result as remarkable. The institute synthesized the vaccine and administered it to Rosie under proper veterinary oversight.
What Changed: The Convergence That Made This Possible
This case did not happen in a vacuum. It sits at the intersection of four converging technological trends that have been building for years.
1. The Plummeting Cost of Genomic Sequencing
DNA sequencing that cost over $100,000 per genome in 2013 now costs approximately $3,000, according to Genome.gov data. This 97% cost reduction is what made Conyngham’s project financially feasible for an individual rather than an institution. Services like those offered by Illumina have driven this compression, making whole-genome sequencing accessible to anyone willing to pay roughly the cost of a used laptop.
2. AlphaFold Went Public
DeepMind’s AlphaFold solved one of biology’s most intractable problems — predicting how proteins fold from their amino acid sequences. Introduced in 2020 and made freely available in 2021, it gave anyone with an internet connection access to protein structure prediction that previously required expensive X-ray crystallography or cryo-electron microscopy. In Conyngham’s pipeline, AlphaFold identified the mutant proteins in Rosie’s tumor and determined which could serve as viable therapeutic targets.
3. Large Language Models as Research Partners
ChatGPT served as Conyngham’s primary interface for understanding biological concepts, interpreting sequencing data, and designing the mRNA sequence. This is a qualitative shift. A non-expert used a general-purpose AI to navigate a domain that typically requires years of graduate training. The model didn’t replace expert knowledge — it bridged the gap between Conyngham’s data engineering skills and the biological domain knowledge he lacked.
4. mRNA Technology Matured
The COVID-19 pandemic accelerated mRNA vaccine technology by decades. Moderna’s mRNA COVID-19 vaccine was developed in under a year in 2020, proving that the platform could move from concept to injection at unprecedented speed. That same underlying technology — encoding instructions for the body to produce specific proteins — is what Conyngham adapted for a personalized cancer vaccine targeting Rosie’s specific tumor neoantigens.
The Competitive Landscape: Who Else Is Doing This?
Conyngham’s DIY approach is dramatic, but it exists within a broader wave of AI-driven personalized oncology that major players are already pursuing.
Moderna and Merck have been running clinical trials on personalized mRNA cancer vaccines since 2023. Their V940 vaccine, combined with Keytruda (pembrolizumab), showed a 44% reduction in melanoma recurrence risk in Phase 2b trials. Five-year follow-up data continues to show durable responses. (Scientific American)
BioNTech, Moderna’s European counterpart, is running its own individualized neoantigen therapy (iNeST) program across multiple cancer types. In China, companies like Stemirna and cloud-based biotech firms backed by Yunding Xinyao are racing to bring personalized mRNA cancer vaccines to market, with some already in clinical trials. (JCN Newswire)
Oracle CEO Larry Ellison has publicly stated that AI-powered personalized cancer vaccines represent one of the most promising applications of artificial intelligence, predicting that within a few years, every cancer patient could receive a vaccine tailored to their specific tumor mutations.
What sets Conyngham’s case apart is not the science — it’s the accessibility. He demonstrated that the tools to do this are no longer locked behind institutional walls.
Related: The Non-Technical Person’s Guide to AI in 2026: What Actually Matters
What This Means for AI Tool Users
If you’re reading this on an AI tools site, here’s why this story matters to you specifically:
AI is no longer just a productivity tool. The same ChatGPT subscription you use for writing emails and debugging code was used to design a cancer vaccine. The ceiling on what general-purpose AI can help accomplish has risen dramatically.
Domain expertise barriers are falling. Conyngham’s case proves that someone with strong data skills but zero domain knowledge can produce meaningful results in a completely unrelated field — if they know how to use AI as a research partner. This pattern will repeat across law, finance, engineering, and other specialized domains.
The real skill is pipeline assembly. Conyngham didn’t build any of the individual tools. He assembled them — sequencing services, AlphaFold, ChatGPT, university lab partnerships — into a coherent workflow. This is increasingly what “using AI well” looks like: knowing which tools exist, how to chain them together, and where human oversight is still essential.
Validation still requires experts. Conyngham worked with the UNSW RNA Institute for synthesis and administration, and went through a 3-month ethics board approval process. AI can accelerate design, but it cannot replace clinical validation, safety testing, or regulatory compliance. Anyone attempting to replicate this approach without proper expert oversight would be taking serious risks.
The Risks and Limitations
This story is inspiring, but it comes with important caveats.
A single case in a dog is not a clinical trial. The 50% tumor shrinkage is promising but has not been replicated in controlled conditions. Mast cell tumors in dogs can sometimes respond to various interventions, and without a control group, it’s impossible to attribute the result solely to the vaccine.
DIY medicine carries real dangers. An incorrectly designed mRNA sequence could trigger dangerous immune responses. The fact that Conyngham worked with university researchers and went through ethics review is a critical detail — this was not a garage experiment.
Regulatory frameworks are not ready. Current FDA and EMA guidelines for AI-assisted drug design are still evolving. The EU AI Act and updated FDA guidance on AI in medicine both emphasize transparency and bias mitigation, but neither fully addresses the scenario of individuals designing personalized treatments with consumer AI tools.
Data privacy is a concern. Genomic data is among the most sensitive personal information that exists. As AI-powered personalized medicine scales, questions about who owns, stores, and has access to this data will become increasingly urgent.
Where This Goes Next
The global AI in healthcare market is projected to reach $187.95 billion by 2030, growing at a compound annual growth rate of 40.6%, according to Grand View Research. Personalized medicine specifically is expected to hit $717 billion by 2025 per Statista projections.
The practical trajectory is clear: end-to-end platforms that integrate sequencing, neoantigen discovery, structure prediction, immunogenicity scoring, and manufacturability checks — all with proper audit trails — will become the next major category of AI-powered health tools. Companies that can package this pipeline into something accessible, validated, and compliant will capture enormous value.
For now, Rosie is alive and chasing rabbits in Sydney parks. And a data engineer with a ChatGPT subscription proved that the future of personalized medicine is closer than anyone expected.
Next Step
Use these pages to keep the decision moving:
- More in AI Chat — Keep researching the same category instead of stopping at one article.
- Open comparisons — Jump into direct matchups and trade-off pages.
- Open tool guides — Use the canonical decision pages for fit, pricing context, and alternatives in one place.