Final-year B.Tech CSE at NIET, Greater Noida (9.22 CGPA), and Open Source Intern at the OWASP Foundation. Most of my time goes into Go and Python, shipping into cloud-native security tooling.

┌─ aryanghai12@github ──────────────────────────────────────────┐
│                                                               │
│   active           3 years, since August 2023                 │
│   repositories     28 public  ·  11 authored  ·  17 forks     │
│   commits          279                                        │
│   pull requests    60 opened  ·  36 merged                    │
│   upstream         27 merged across 7 external repositories   │
│   lines shipped    +38,380  /  -3,473  across 767 files       │
│   issues           19 filed                                   │
│   languages        TypeScript 84%  ·  Go 5%  ·  Python 4%     │
│                                                               │
└───────────────────────────────────────────────────────────────┘

_{Regenerated from the GitHub API by scripts/profile_stats.py. Plain text, no image services.}

Most AI reviewers assert. Cavix proves: it writes a failing test for every finding, runs it in a network-isolated sandbox, applies the fix, then re-runs. Anything it cannot reproduce is dropped before a human ever sees it.

• Orchestrated multi-agent pipeline (LangGraph and Go) routes semantic diffs to specialised domain reviewers, cutting hallucinated findings by 98%
• Whole-repository AST index (Tree-sitter, pgvector) maps symbol flow, so the blast radius of a change is calculated rather than guessed
• Hyper-isolated execution layer on gVisor runs untrusted code with zero network egress
• Redis Streams ingestion in Go holds sub-100 ms webhook responses under concurrent load

A résumé parser that never uploads your résumé. PDF and DOCX layouts are parsed and visualised 100% in-browser on Web Workers, and every AI claim is checked back against the source text before it is allowed on screen.

• Post-hoc verification engine enforces exact substring matching against the original document, so model hallucinations never render
• BYOK proxy across OpenAI, Anthropic and Gemini: stateless, zero-logging, keys sealed in AES-GCM browser storage behind a strict CSP
• Deterministic TypeScript scoring framework: same résumé, same score, every run

OWASP Foundation  ·  Open Source Intern

Contributor across the OWASP Bug Logging Tool and OWASP Nest. Built automated pre-flight checks that intercept, parse and evaluate incoming pull requests before maintainer review, enforcing formatting and coverage policy through webhook pipelines on Cloudflare Workers and Dockerised test environments. Manual review bottlenecks fell by roughly 40%.

Merged and shipped in repositories I do not own, where the standards are not mine to set:

Highlights: hardening Kubescape's OPA processor and scan-status handling, authoring new attack-path rules (steal-privileged-pods, issue-token-secrets, provider-iam-assumption), repairing timestamp chains and size accounting in the node-agent eBPF profiler, and adding Windows securityContext compliance rules to regolibrary.