I build backend systems that make security work faster and more reliable: serverless services, API infrastructure, and AI tooling that has to be trustworthy, not just impressive in a demo.
B.Tech CSE at AKTU (graduating 2027) · Google Cloud Certified in Cybersecurity and Cloud Computing Foundations.
AgentGuard — reliability & evaluation for tool-using LLM agents
Most agent demos answer "did it work once?" AgentGuard asks the harder question:
How reliable is a tool-using LLM agent when the same task is repeated, and when controlled failures are injected?
- A small, fully-owned evaluation system that tests one reference agent, in one controlled domain, end to end.
- Every metric traces back to a real stored run.
- Every LLM-judge score ships with a human-agreement calibration rate, never presented as ground truth.
- Not a Langfuse clone, not a DeepEval replacement, not a multi-agent framework.
BLT-Toasty — AI triage & responsible-disclosure assistant for OWASP BLT
Built for open-source security teams. Redacts PII, summarizes reports, and flags duplicates (~70% of submissions) in under 2 seconds, cutting manual review from 10–15 minutes to near-instant.
Cloudflare Workers · Durable Objects · D1 · Vectorize · GitHub Actions
| When | Role | Impact |
|---|---|---|
| Jun – Aug 2026 | Software Engineering Intern, OWASP BLT | Architected BLT-Toasty's serverless backend on Cloudflare Workers, cutting triage time by 30% with zero PII leakage. |
| Feb 2026 – Present | Open-source Contributor, OWASP BLT | 25+ merged PRs refactoring Python microservices and hardening API security. |
Also: Unit testing · CI/CD automation · Data structures & algorithms · Cloud & API security fundamentals
Open to Backend Engineer · AI Security Engineer · Python Developer · Software Engineer roles.
📧 ojaswa717@gmail.com · 💼 LinkedIn
$ sudo brew install coffee --for ojaswa
==> Sponsor on GitHub: https://github.com/sponsors/ojaswa072