Hi, I'm Shivam
Designing and building production software, with help from a small team of AI agents.
Freelance, or the right full-time role.
I run HireRank as a team of one. I like building things this way. Mechanical keyboards, LEGO worlds, software. Same instinct.
I started HireRank in late 2025 to test a small thesis: hiring engineers from resume keyword density is broken. The platform scores resumes against job descriptions, rewrites the weak parts, and verifies real-world output through GitHub, Codeforces, and LeetCode. Eleven releases in. Around a hundred daily users. Paying subscribers since the second month. Every line is mine, and so is the on-call.
Spent a summer at Tata going through years of factory floor safety data. Incident reports, PPE compliance, near-miss logs. Found that violations clustered in patterns nobody had charted before. The operations team picked up several of those findings for the next safety protocol revision. NumPy, TensorFlow, and Matplotlib did the heavy lifting.
Two years running a ten-person spirits retail team in heavily regulated territory. I learned production discipline before I learned production code: daily compliance filings, licensing paperwork, audit trails, and the actual cost of getting any of it wrong. Worked with the Department of Excise and Prohibition to redesign the licensing workflow. End-of-day close got shorter. The auditors had less to complain about.
Data Science specialization · Kattankulathur, TN
Mathematics, Science, Computer Science · Hazaribagh, JH
I run a studio of AI agents, with one human at the keyboard.
Each agent has a job. Claude talks architecture. Claude Code writes the implementation. CodeRabbit reviews every PR before I see it. Codex, ChatGPT, and Gemini come in as second opinions when I'm stuck. Antigravity codes in the background, Firecrawl pulls context, BrainTrust runs evals, and Stitch sketches the UI.
This is the kind of work that used to employ a designer, two developers, a QA engineer, and a few consultants. The team is smaller now. The discipline isn't.
The agents have their toolkit. Mine is just two windows: Ghostty for the terminal, Zen Browser for everything else.
The agents do a lot. I direct. The accountability is mine when something breaks.
Selected projects & collaborations.
A platform for hiring engineers from what they build instead of what their resume claims. Scores resumes against job descriptions, rewrites the weak parts, and verifies real output through GitHub, Codeforces, and LeetCode. Live at hirerank.xyz with about a hundred daily users.
Audio in, queryable knowledge out. AssemblyAI handles transcription, Gemini does the summaries and questions, and MongoDB Atlas Vector Search lets you find a moment from any lecture by meaning instead of keyword.
Rainfall classifier on 145K rows of Australian Bureau of Meteorology data. Seven categorical features, a 78/22 class imbalance, and a quiet surprise at the end: the logistic regression baseline beat the deeper TensorFlow network.
A native Mac app that reads your design system from the source files. Colors, typography, spacing, shadows, motion. It renders the tokens live as a real, working interface. Buttons, cards, inputs, navigation, all built from your own tokens. When someone changes one, Calden quietly snapshots what shifted and shows you the diff inside the components themselves. A calm window into your design system.
