Build in Public

Six Landing Pages, Each Mapped to a Number

A PageSpeed Score of 76 and Five Separate Bugs Behind It

The Logs Were Missing. So Was the Deploy Pipeline's Connection to the Server — and Nothing Said So.

I Added Analytics and Doubled the Homepage's JavaScript. I Caught It Before Anyone Else Did.

A Free Trial Behind a Flag Nobody Could Flip Without Logging In Themselves

A Secret From the Second Commit, Found Five Months Later: Finding a Hardcoded JWT Secret in a Security Audit

A security review of MockEvalio found admin auth silently broken, a cross-tenant data leak, and a hardcoded key that traced straight back to the project's second commit, five months earlier.

I Verified the Payment Flow End-to-End. The Buttons Still Didn't Work: A Frontend Bug Hiding Behind a Passing Backend Test

I tested MockEvalio's new Stripe checkout end-to-end — real test card, real webhook signature. Three hours later, the buttons meant to get someone there were silently broken.

I Built a System to Grade My AI Grader. I Never Gave It Anything to Grade Against: The Missing Benchmark for an AI Interview Evaluator

I built a complete system to measure whether MockEvalio's AI evaluator is accurate — comparison logic, deviation thresholds, a calibration worker, an admin dashboard. I never built the one piece that would let it run.

Three Layers of AI Infrastructure, Thirty-Three Minutes Apart: Debugging a Multi-Provider AI Orchestration Stack

In thirty-three minutes, MockEvalio went from calling Groq directly from five different services to a centralized orchestrator with versioned prompts and cost-aware multi-provider routing. What built on what, and what's never been touched since.

₹399 Became Three Different Prices, and I Don't Know Why: A Pricing Bug in Razorpay Subscription Tiers

MockEvalio's single ₹399 plan became three separate prices within nine hours of its first real payment. The change is fully on the record. The reasoning isn't, and I looked.

Full Observability Was Too Loud By the Next Morning: Taming Alert Noise in a New OpenTelemetry Setup

I built full OpenTelemetry coverage into MockEvalio over one night. By morning, part of what I'd just built was already too loud to keep as-is.

I Kept Two Job Providers, Then Deleted Them Nineteen Minutes Later: A Job-Matching API Integration Reversal

I committed a plan to keep two job-search providers as optional extras. Nineteen minutes later, I committed the opposite. The commit history proves it happened. It doesn't say why.

LLM Output Repetition Has Three Independent Causes, and Fixing One Isn't Enough

An LLM that keeps reaching for the same phrasing usually has three separate causes stacked on top of each other. Fixing one rarely fixes the output.

The Follow-Up Question That Always Started With "What Would You Sacrifice"

MockEvalio's interview AI kept reaching for the same follow-up phrasing. The fix wasn't one change — it was three, aimed at three different parts of the problem.

The Login Button That Also Filled In Your Profile

Two of MockEvalio's three OAuth logins just logged you in. The third also read your job title and skills out of your headline. Here's the mechanism, and what's still unexplained about the decision.

The Resume Feature I Rebuilt Five Times in 34 Hours

MockEvalio's resume scorer worked within hours. Turning it into resume intelligence took five more rebuilds in a day and a half — and the record only tells me what changed, not why.

Our First RAG Pipeline Failed on Day One

MockEvalio's first RAG pipeline shipped and broke on the same day. What the evidence shows about the failure, the fix, and what the fix didn't prove.

The Second Commit

Two hours after MockEvalio's repository existed, a complete application did too. What that actually proves, and what it doesn't.