About Graphify

Graphify is a code-intelligence control plane for the age of AI-written software. We build a live knowledge graph of a codebase, an AI PR-review layer grounded in that graph, and a differential formal verification engine that answers the question tests and LLM reviewers cannot: does this change actually preserve behavior, for every input in the checked domain.

We run on-prem and in air-gapped enterprise environments, where "trust me" loses to "here is the proof." We are a Y Combinator company (S26) with production usage across regulated and large-scale engineering teams.

Coding agents now generate more code than any team can read. The bar has to move from review toward verification. That is the problem this role owns.

The role

You will advance the core of our formal verification engine: the part that takes two versions of a real function (the pre-change version is the specification) and either proves them behaviorally equivalent, produces a concrete distinguishing input, or abstains honestly. You will push sound coverage up across languages, raise the rate at which we can drive real inputs into real code, and keep every verdict sound by construction.

What you will do

- Develop and improve the differential (relational) verification engine that decides whether an

AI-generated or human change preserves behavior, across Python, Go, Rust, Java, C, C++, TypeScript, and COBOL, using SMT (Z3), property-based differential testing, and trace-carving from a repository's own test suite.
- Formalize the operational semantics of real-world languages for our sound tiers, build verification-condition generation (vcgen) and sound static analyses over the code knowledge graph,



and extend the CEGIS and proof-search loops that discharge them.
- Attack the capture-rate problem: design input-synthesis and receiver-construction techniques (feedback-directed generation, call-site mining, constructor synthesis) that feed the unchanged sound oracle, turning honest abstentions into proven or refuted verdicts.
- Design and run experiments on real enterprise PRs, OSS corpora, and mutation benchmarks: measure sound-verdict yield, non-vacuity, capture-rate, and mutation score against production telemetry, and turn findings into shipped improvements.
- Build infrastructure to run verification at scale in the PR path and inside air-gapped deployments: tiered proof search orchestration, per-language front-ends, proof-carrying certificates that a third party can re-check, and integration with the code graph and hosted worker.

Minimum qualifications

- PhD in programming languages, formal methods, or a related area, or equivalent practical experience.
- 4 years of experience with real-world software verification and bug-finding.
- 3 years of experience with proof assistants (Lean, Rocq, or similar) or SMT solvers (Z3, CVC5) and model checkers (CBMC, JBMC, Kani).

Preferred qualifications

- Experience formalizing the semantics of real-world programming languages and building sound analyses on top.




- Experience with differential or relational verification, equivalence checking, regression verification, or translation validation.
- Experience with automated test and input generation (feedback-directed, SBST, concolic, property-based) applied to object-oriented or framework-heavy code.
- Track record of publications at top venues (PLDI, POPL, CAV, OOPSLA, ICSE, FSE, ISSTA) or equivalent open-source impact.
- Comfort shipping verification into a production, latency-bounded, secret-secure pipeline.

Why this role is different

- Spec-free by design. The old code is the specification, so we deliver sound guarantees on real enterprise codebases without asking anyone to write a formal spec first.
- Sound only. Every accepted verdict is backed by a decision procedure or a reproduced counterexample, gated for non-vacuity. We never ship a guessed pass.
- Production and on-prem. Your proofs run in the PR check and inside air-gapped environments, on code that matters, at team speed.

Compensation and benefits

Competitive salary plus meaningful founding-team equity, and benefits. Bands are set by job-related skills, experience, and location; we will share specifics early in the process.

How to apply

Send your CV and a short note (or a link to work you are proud of: a solver-backed tool, a verified analysis, a hard proof) to [email protected]

- Graphify is an equal opportunity employer. We consider all qualified applicants without regard to race, color, ancestry, religion, sex, national origin, sexual orientation, age, citizenship, marital status, disability, gender identity, or veteran status.

📌 Research Engineer, Verified Code Generation (India)
🏢 Graphify Labs (YC S26)
📍 India