Hire for Pull Requests, Not for Pedigree

A candidate's merged code reveals more than their resume's brand names ever will.

The hook

We dropped a candidate into a staging environment with a real memory leak in a Node.js service. IIT Bombay, CS. Two years at Google. Clean GitHub profile with a pinned repo and no commits in nine months. After 25 minutes with the heap profiler, they asked if they could "talk through the approach instead." The next day, someone from a college I had never heard of found the leaking closure in eleven minutes and opened a PR with a regression test before the hour ended.

महारासी शिवे · If Touching an Outcast Enrages You

महारासी शिवे । कोपे ब्राह्मण तो नव्हे ॥
ज्याचा संग चित्ती । तुका म्हणे तो त्या जाती ॥
You are defined by your conduct, not your label. You belong to the caste of your behavior, not your birth certificate.

What I keep seeing

Our hiring panel nearly passed on the second candidate in the resume screen. The recruiter had flagged them as "non-target school, no brand-name experience." We only interviewed them because an engineer on the team recognized one of the open source projects they had contributed to. Two hundred and seventeen merged PRs across four projects: two Rust crates, a Kubernetes operator, and a PostgreSQL extension. Without that accident of recognition, we would have filtered out the best candidate we interviewed that quarter.

I have made this mistake myself, earlier in my career. I once passed on a candidate because their resume did not have a single company I recognized. They went to a competitor. Built their real-time pipeline. I spent six months wondering what happened.

The mechanics

Resume screening is a classifier, and like all classifiers it has a false-positive rate and a false-negative rate. When you filter on school name and company brand, you optimize for precision at the expense of recall. The base rate matters: roughly 23 IITs graduate around 16,000 engineers a year. Over 3,000 other engineering colleges graduate more than 1.5 million. Even if the hit rate at IITs is ten times higher, the absolute number of strong engineers outside the system dwarfs the number inside it. You are fishing in the wrong pond.

A merged PR on a maintained open source project means the candidate read someone else's codebase, identified a change worth making, wrote code that passed CI, responded to review feedback, and got a maintainer to approve the change. That sequence (read, write, respond, ship) is the actual daily work of a professional software engineer. A degree from a top school means the candidate passed exams. These are not the same skill.

Where Tuka comes in

Tukaram's couplet is two lines long and leaves no room for negotiation. If a Brahmin rages at contact with an untouchable, he is not a Brahmin. Full stop. The label is void. What makes you what you are is "ज्याचा संग चित्ती" (the company your mind keeps): the actions you actually perform, not the title printed on your certificate.

This is not a metaphor loosely applied. It is the literal structure of the hiring problem. The industry assigns a label (IIT, FAANG, "target school") and treats the label as a reliable proxy for ability. Tukaram says the proxy is worthless if the conduct does not match. A candidate with the right label who cannot debug a running process is not a senior engineer. A candidate with no label who ships tested fixes to production open source projects is. You belong to the caste of your behavior.

What I would actually do

Replace resume-based screening with a structured work-sample review. Before any interview, ask candidates to submit three to five merged PRs, open source or from a previous employer's public repos. If they have no public code, give them a take-home that simulates a real PR: a repo with a bug, a failing test, instructions to open a pull request with a fix. Evaluate it the way you would evaluate a colleague's PR: code quality, commit messages, test coverage, response to review comments. Drop school name and company name from the resume before it reaches the hiring panel. The engineer who debugged a memory leak in eleven minutes and shipped a regression test did not learn that skill in a lecture hall. They learned it by doing the work. Screen for the work.

Chetan Dhandal