Ayush Kumar

I’m a software engineer-intern and independent LLM security researcher focused on building reliable backend systems and improving real-world security. I enjoy turning requirements into production-ready features, designing modular components, and raising system quality through careful debugging and testing.

At Carboledger, I worked on a multi-tenant compliance platform with complex domain workflows. I merged 65 PRs in 16 weeks, converted requirement cards into extensible backend features using design patterns and modular architecture, and owned critical bug identification and resolution across cross-version testing and regression cycles.

Earlier, at Fast Conversion eSolutions Private Limited, I improved web performance and API efficiency as a Web Development Intern. I reduced API latency using Redis session caching and async Express middleware, then refactored 12 REST endpoints to eliminate N+1 queries and add connection pooling—cutting down average database query time.

I also pursue security research through independent LLM security work. I identified and responsibly disclosed a critical prompt leakage vulnerability in a Claude-based code review platform (CodeVerdict), providing root cause, reproduction steps, and actionable remediation; I was recognized for report quality and security depth. In parallel, I build ML/CV solutions like a CDM-based fake image detection pipeline and a Flask-based toddler autism prediction API achieving strong evaluation results, guided by careful end-to-end experimentation.