I’m Justin Johnson. I Build Things.
justinhjohnson.com
Twenty years in biotech. Executive Director at a global pharma company. Strategy meetings, stakeholder management, roadmaps. Meaningful work that helps get medicines to patients.
But that’s not who I am.
I’m a builder. Always have been.
The Day Job
I lead Data & AI Platforms for oncology research at one of the world’s largest pharmaceutical companies. My team builds the infrastructure that lets scientists discover and develop cancer treatments faster. We’ve deployed agentic AI systems that compress weeks of analysis into hours. We’ve built data platforms that make clinical trial information instantly accessible. We’ve created tools that thousands of researchers use daily.
It’s the best job I’ve ever had. Front-row seat to AI transforming drug discovery. Real impact on real patients.
But here’s the thing about leading teams in a big organization: you spend most of your time in meetings, not in code. You’re enabling others to build, not building yourself. That’s the right trade-off for the work we’re doing. But the itch doesn’t go away.
The Night Shift
So I build at night. Weekends. Early mornings before anyone else is awake. On planes. In hotel rooms.
Not because I have to. Because I can’t not.
Over the past 18 months, that compulsion has produced something I never planned: 34 distinct projects spanning enterprise platforms, research experiments, trading systems, robots, and a growing content operation.
Navari: An agentic AI platform with 550+ users and 2900% growth
OncoVLM: Oncology-specific multimodal models achieving 92.4% on PubMedQA
Agora Quantum: A 9-agent autonomous trading system scanning 8,000+ securities daily
ARIA: A research agent that runs 24/7, generating novel hypotheses while I sleep
Dendrite: A Rust-based inference engine with 10,000x faster branching than vLLM
Each project taught me something. Each failure refined my approach. The AI collaboration skills compounded. What used to take weeks now takes days. What used to require a team now requires me and a well-crafted prompt.
The Philosophy
I believe we’re living through the most significant shift in how knowledge work gets done since the invention of the spreadsheet. Maybe bigger.
I call it the 1:N effect. One person, properly equipped with AI collaborators, can produce output that previously required teams. Not because AI replaces human judgment — it doesn’t — but because it eliminates the friction between having an idea and making it real.
Think about what used to stand between “I have an idea for an app” and “here’s the working app”:
Learning the framework
Writing boilerplate
Debugging syntax errors
Googling Stack Overflow
Context-switching between documentation tabs
AI collapses all of that. The gap between imagination and implementation shrinks to nearly zero.
This isn’t about working harder. It’s about leverage. The same effort produces 10x the output when you’re not fighting the tools.
The Evidence
I wanted to capture this somewhere. Not a resume — resumes are dead. Not a LinkedIn profile — those are for recruiters. Something that shows the work, the connections between projects, the patterns that emerged.
So I built justinhjohnson.com.
The site is an interactive archive of 18 months of building. Every project is a node in a network graph. Click one and see what it does, why I built it, what it connects to. Filter by framing to see different slices: enterprise work, research experiments, personal tools, content systems.
There’s a story mode for people who prefer narrative over exploration. A timeline view showing how projects evolved. A stack view mapping everything to the underlying infrastructure.
The meta-joke is that the site itself was built in two days using the same AI-assisted workflow it documents. Claude wrote the code. Gemini generated the images. I curated and edited. The recursion is the point.
What I’m Exploring Now
The frontier keeps moving. Here’s where my attention is:
Agentic systems that take action, not just answer questions. The shift from “AI as search engine” to “AI as colleague” is the most important trend in the field. Navari embodies this — it doesn’t tell scientists what to do, it does it.
Autonomous research agents. ARIA runs continuously on my DGX, generating novel hypotheses at the intersection of AI and biotechnology. It’s a living research organism that compounds knowledge over time.
Small language models and training. I’m obsessed with what small, focused models can do. Training domain-specific models from scratch using LoRA and reinforcement learning techniques. The bet: a 4B model trained on the right data beats a 70B model trained on everything. OncoVLM’s 92.4% PubMedQA score is early evidence.
The content flywheel. Every experiment becomes a blog post. Every blog post becomes social content. The publishing pipeline is automated end-to-end — from DGX training logs to Substack to LinkedIn.
The Invitation
If you’re building with AI, I want to hear about it. The community of people pushing these boundaries is still small enough that we can learn from each other.
Find me:
Blog: rundatarun.io — technical deep dives and building-in-public stories
Portfolio: justinhjohnson.com — the interactive archive
Twitter/X: @BioInfo
GitHub: @BioInfo
LinkedIn: Justin H. Johnson
The future belongs to builders. Let’s build it together.
By day, I help discover cancer treatments. By night, I prove what one person can do with AI. The two aren’t as different as you’d think.