Three months, idea to production.
KaizenShots is our own product. An AI mentor for photographers that reads a photo and returns a critique, scores, an editing plan, and an honest read on whether it is worth publishing. It started as an empty repository in mid April. It went live to the public in mid July. This page is how.
Twelve weeks. One builder.
No agency, no offshore team, no six-month discovery phase. The interesting part is not that AI helped. It is how much product one person could carry with it.
Mid April to mid July
Twelve weeks from an empty repository to a product the public can sign up for.
One builder
Design, architecture, development, and unit tests. A DevOps engineer handled deployment. A QA engineer tested it by hand.
Live and open
Running in production today, with free and paid plans. Not a demo, not a pilot.
Not a wrapper on a model.
Most AI side projects are one prompt and a text box. This is the whole product, and the scope is the point.
The critique pipeline
Every photo runs three sequential model passes. Feedback and scores, then professional analysis and publication fit, then an editing plan with a preset and masking.
Three model providers
OpenAI, Anthropic, and Google sit behind one interface. The model can be switched from the admin console without a deploy.
A real job queue
A background worker with a concurrency ceiling set at runtime, stuck-job recovery, and restarts that do not lose work.
Accounts and payments
Email and Google sign-in, subscriptions and one-off purchases through Razorpay, with signature-verified webhooks.
An admin console
Models, prompts, settings, and analytics are all run from inside the product, not from a config file.
Everything else a product needs
Image moderation, presigned uploads, PDF export, transactional email, error monitoring, three deploy environments, unit and end to end tests.
What AI did, and what it cost.
The useful part of building your own product is that you cannot lie to yourself about the tooling afterward.
Agentic tooling wrote most of the code.
Whole modules, migrations, and tests came out of agentic workflows. That is how one person covered the ground in twelve weeks.
It does not decide anything.
Architecture, data model, and product calls stayed human. The tool is fast, not senior. Treating it as senior is how projects rot.
It over-engineers, constantly.
Left alone it builds more than the problem needs. Abstractions nobody asked for, options nobody will use. Holding it back was a daily job, and it is the part most people do not mention.
The build fails when a prompt drifts.
Prompts are recorded in a manifest and checked in CI. If one changes without being declared, the pipeline stops. AI code needs tests the same way any other code does.
Boring where it counts.
Nothing exotic. A stack a new engineer can read on day one, because someone has to own it after us.
LIVESee it yourself.
KaizenShots is open to everyone. Sign up, upload a photo, and judge the output.
Visit KaizenShots ↗Same method, your roadmap.
Whether you are a founder taking an idea to launch or a product team that needs senior hands, tell us what you are trying to ship.