Vaibhav Rana
Engineer by training, product leader by trade, builder by compulsion. And for the last eighteen months, an AI operator with my own name on the results.
I started as a mechanical engineer: a BEng at Cardiff, then an MPhil in Industrial Systems, Manufacture and Management at the University of Cambridge. Manufacturing teaches you a habit that never leaves: a process you cannot measure is a process you cannot trust. Everything BlooSprout does about verification traces back to that.
From there I moved into software and spent over a decade leading product, most recently heading product at a venture-backed e-commerce SaaS, where my remit runs from roadmap to launch to the analytics that say whether any of it worked. Along the way I kept building things of my own: a no-code education venture, a remote-work company, a property business I still run today, and a free AI workshop I taught at my local museum because someone had to tell small businesses the truth about this stuff. Some of those ventures worked. Some I closed. Both taught me more than the job titles did.
And for the last eighteen months I have built and operated AI systems daily: marketing engines, reporting pipelines, personal assistants, verification harnesses. Not advising on them from a distance. Running them, with my own name on the results.
The two-day lesson
In 2025 I spent months hand-building a personal AI assistant with the standard framework stack. It demoed well. It never shipped.
In 2026 I rebuilt the same idea with a different method: a written brief the AI cannot drift from, machine-checkable definitions of done, and verification gates on every change. It went from first commit to production in two days, replaced a paid subscription the next day, and has run my real inbox and calendar ever since.
The AI models barely changed between those two attempts. The method changed everything. BlooSprout exists to sell that method.
How I work with AI
Direct
I brief AI the way you would brief a senior hire: the goal, the constraints, what the project is NOT, and the one metric that decides success. Most stalled pilots die at the brief, not the build.
Orchestrate
Real work needs more than one pair of hands, so I run teams of AI agents with different jobs: builders, reviewers that check the builders, and judges that settle disagreements. The strongest model does the hard thinking; cheaper ones do the mechanical checking. That discipline is also why the economics work.
Verify
Nothing ships on trust. Every system I run answers to tests, scored eval runs, and a monthly scorecard that shows failures as plainly as wins. This site publishes its own numbers on the receipts page, including the ones I would rather improve before showing you.
Codify
When something works once it becomes a system: a written playbook, a quality gate, a scheduled routine. That is why the work compounds instead of resetting with every project, and why you own everything when I hand it over.
What that looks like run hot
The marketing engine behind a property platform I run shipped in one night: 101 commits while I slept, every one gated on the test suite, and the platform holds 1,400 green tests today. This site's blog, free tools, and search plumbing are maintained by the same machinery. And because every number needs a receipt here, each figure in this paragraph sits in a public claims registry with the command that re-derives it.
If that sounds like the way you want AI run in your business, start with the functions we operate and a free 15-minute audit to find which one comes first.