
In My Own Words
I spent 18 months building something I wasn't sure would work: an AI system that lets researchers query the Human Protein Atlas (HPA) database using plain English instead of complex database queries.
The result? A multi-agent system that achieved 93.6% accuracy against ground truth data, with zero hallucinations across 12 benchmark tests. Scientists can now ask questions like "find liver-specific proteins with validated antibodies" and get verified results in minutes instead of hours of manual searching. Read: https://axonagentic.ai/blog/ai-natural-language-human-protein-atlas-18-month-journey
This project taught me that the hardest part of AI isn't the technology—it's building systems that scientists can actually trust. Every claim the system makes is cross-validated against HPA's public API. Every result includes the methodology to reproduce it independently.
Before founding Axon Agentic, I spent 12+ years in digital transformation at Thermo Fisher Scientific, EMD MilliporeSigma, and Ogilvy/WPP. I learned how life sciences companies actually work—the workflows, the validation requirements, the gap between what vendors promise and what researchers need.
I don't have a PhD. What I do have is 18 months of building, breaking, and rebuilding until something worked. The stack spans backend APIs, cross-platform clients, and a distributed multi-service architecture. The AI layer coordinates 8 specialized agents using large context windows, routing queries with 100% accuracy.
The data infrastructure—analytical databases, knowledge graphs, and vector search—handles 273 million tissue expression records and 401,855 commercial antibody products.
I'm currently exploring opportunities where I can apply this experience: building AI systems that require verification, working with complex biological or scientific data, or helping teams move from AI pilots to production.
If you're working on similar problems—or just want to talk about what actually works in production AI—I'd enjoy connecting.