The Second Commit

At 20:29 on March 9, I created a repository called mockevalio. The only thing in it was a README with one line: # mockevalio. Two hours and twenty minutes later, at 22:50 the same night, the second commit landed — and it wasn't a small addition. It was a working ASP.NET Core 8 backend with a real Clean Architecture split, a React frontend with nine pages already built, a Postgres schema with eight related entities, thirteen API endpoints across six controllers, JWT authentication, and a Docker Compose setup ready to deploy. Nothing in between. Empty, then running, in one jump.

That gap is the actual starting point for this log — not because it's impressive, but because of what it does and doesn't prove.

What existed two hours later

The entity model from that second commit already connected a user to interview sessions, sessions to answers, answers to AI-evaluated results, and users to subscriptions — the shape of "practice an interview, get scored, pay for more of it" was there from the first working version, not added later. Resume scoring wasn't in it yet; that came two days after. The pricing model was a single field on a Subscription record, defaulting to "Free" with "Pro" as the only paid option — matching a UI plan document that shipped in the same commit, which described a ₹399/month Pro tier and a hero headline: "Ace Every Interview with AI-Powered Evaluation." The only AI provider configured was Groq. Razorpay's payment keys were already wired in, empty, waiting to be filled — before there was any code that actually charged anyone.

How this actually got built

I didn't write that second commit by hand, line by line, in two hours twenty minutes. It came out of Cursor, an AI-assisted development tool — and that's not a one-time detail about how the project started. Across the project's history since, roughly half of all commits — 116 of 226 — carry the same tag, including work months later like a full rewrite of how the app routes its AI calls. AI-assisted development wasn't how MockEvalio got scaffolded before the real work began. It's been part of the real work the whole time.

What that doesn't mean: that the tool made the decisions. The architecture, the bounded contexts, the security model, the AI evaluation approach, the infrastructure, the testing, the trade-offs, the product direction — those are mine, and stay mine, regardless of which editor generated which lines. I'm not going to claim I typed every character, and I'm also not going to pretend the tool ran the project.

What two commits don't tell you

The gap between commit one and commit two proves a timestamp, not a story. I don't know, and I'm not going to invent, whether that two hours twenty minutes was one sitting or several. I don't know why the Pro price came out to ₹399 specifically — a deliberate number or just what came out of the planning pass. I don't know why Groq was the first AI provider I reached for, ahead of a switch to OpenAI for embeddings that happened the very next day. And the actual reason I decided to start this at all — the real trigger, if there was a specific one — isn't something the repository can tell me either. I'm leaving all four of those open rather than filling them in with something plausible.

The actual question

Producing a working application, it turns out, is now the easy part. I had a product hypothesis — an AI interview-prep platform that does more than ask questions, one that gives evaluation someone can actually use to improve — and AI-assisted development let me get a first version moving in an evening. The interesting question was never whether I could produce a working application. It was whether I could turn that starting point into something genuinely useful, reliable, secure, and worth paying for. That's a different, much harder question, and two commits don't answer it.

Five months and 226 commits later, that question is still open. I've built a lot since that second commit — more than fits in one post, and more than belongs in one. What I actually want to know, over the next 150 days, is whether the thing I hypothesized on March 9 holds up: whether it's useful, whether the AI evaluation is actually good and not just convincing, whether it's reliable, whether it's secure, whether the architecture holds at real scale, whether anyone is willing to pay for it, and whether the answers to any of that mean the product needs to become something different than what I originally sketched out.

I still don't know. That's the actual starting point.