Built by engineers who watched good code review fail at scale
DevPulsr is a bootstrapped product from New York. We started building it in 2021 because we lived the problem firsthand — and found nothing in the market that solved it precisely enough.
Why we built DevPulsr
In 2021, Raj Sundaram was an engineering lead at a fintech startup in Lower Manhattan, managing a 14-person team that processed high-stakes payment flows. His engineers were careful. They wrote tests. They reviewed each other's code. Production still had incidents — not from negligence, but from the inherent limits of human review at PR volume.
A logic inversion in a discount calculation. An API field renamed without a deprecation path. A test suite with 88% line coverage that didn't cover the one branch that mattered. These weren't linter issues. They were semantic drift — code that was structurally valid but contextually wrong — and no tooling in 2021 caught them reliably at PR time.
Raj started building what became DevPulsr as a side project with two colleagues from his NYU Tandon graduate program: Mara Osei, who was doing PhD-level research into code representation learning and semantic program analysis at an NYC AI lab, and Tomas Varga, a backend engineer who'd shipped production static analysis pipelines at two infrastructure companies in Prague and New York.
The first version was a Python script that posted GitHub comments. Then a webhook. Then a proper GitHub App. By early 2024, a dozen teams were running it on their PRs — none of them paying, most of them asking for more. Raj left his job in March 2024 to work on DevPulsr full-time. Priya Nair, who'd spent five years building developer productivity tooling and had seen the same problem from the other side, joined that summer. The four of them have been at it since.
Our operating principles
Specificity over noise
A finding that's 90% accurate but precisely targeted is more valuable than 200 low-signal warnings. We optimize for reviewers trusting the output, not tuning it out.
Code privacy by design
Your source code is your competitive advantage. We designed our infrastructure so no code is ever retained, persisted, or used to train models. This isn't marketing copy — it's the architecture.
Teams, not individuals
Velocity metrics and review analytics should make teams better, not rank individuals. We deliberately don't build leaderboards. We surface team-level patterns instead.
Fast or worthless
Analysis that takes 8 minutes defeats itself — developers move on. Our P50 analysis time is under 90 seconds. Speed is part of the product, not an afterthought.
Bootstrapped discipline
We're not growing toward an acquisition. We're building a sustainable business where every feature earns its place by genuinely helping the teams who rely on it.
Augment, don't replace
DevPulsr is a collaborator in your review process, not a replacement for human judgment. The best reviews happen when AI catches the mechanical and humans focus on the architectural.
Who we are
Raj Sundaram
CEO & Co-founder
Former engineering lead at a fintech startup in Lower Manhattan. Watched good engineers ship regressions at review time for three years before building DevPulsr to solve it. NYU Tandon MSCS, 2017.
Mara Osei
Head of Research & Co-founder
Machine learning researcher specializing in code representation learning and semantic program analysis. Former research engineer at an NYC AI lab. NYU Courant PhD candidate (ABD).
Want to work on this problem with us?
We're small on purpose and not always hiring — but when we are, we're looking for people who write code they're proud of.