AI infrastructure engineer and security researcher. I build production LLM systems, find and responsibly disclose real vulnerabilities, and evaluate AI models for safety and correctness.
I'm an AI infrastructure engineer and security researcher with hands-on experience across production AI systems, application security, and AI model evaluation.
My work spans LLM orchestration and multi-model routing infrastructure, including building an LLM gateway that unifies calls across 750+ models (Llama, OpenAI, Claude, Google, and open-source providers) behind a single API, with real-time cost and latency-aware routing, governance features like immutable audit logging and virtual key management, and support for cloud, hybrid, and self-hosted deployment. I've also built enterprise AI workspace tooling with knowledge-base integration, prompt libraries, and admin controls.
Separately, I find and responsibly disclose real security issues in AI/ML systems. My work includes identifying unsafe deserialization, remote code execution paths, path traversal, and authentication weaknesses in real, actively-used open-source libraries, with published proof-of-concept demonstrations and maintainer-accepted fixes. Recent verified work includes discovering and responsibly disclosing security issues in an open-source ML feature store, a model serialization library, and an ML deployment tool, all independently verified and published as official GitHub Security Advisories (github.com/sahilempire).
I also evaluate and stress-test AI models for correctness, safety, and reliability, reviewing AI-generated responses for accuracy and instruction-following quality, designing prompts that surface edge cases and reasoning failures, and systematically testing AI systems to identify failure points.
My broader technical background covers Python, Java, JavaScript, TypeScript, C, C++, and SQL, with strong skills in technical writing and clear documentation. I work carefully, verify my findings thoroughly, and communicate clearly throughout every engagement.
Available for AI infrastructure development, security audits, model evaluation work, or general software engineering, one-off projects or ongoing work.
Work Terms
Time zone: India Standard Time (IST, UTC+5:30). I'm generally available for real-time communication during Indian business hours and can flex for overlap with US/EU clients when needed for kickoff calls or urgent issues.
Communication: I prefer written communication through the platform for task details and requirements, with video/voice calls available for kickoffs, scoping, or complex discussions. I respond to messages within 24 hours, typically much faster.
Payment: I work with Guru's SafePay for both hourly and fixed-price engagements. For fixed-price projects, I'll provide a clear scope and milestone breakdown before starting. For hourly work, I track time transparently and provide regular updates.
Process: I ask clarifying questions upfront to make sure I understand the actual requirement before starting, this saves time for both of us. I provide progress updates on longer engagements and flag blockers early rather than going quiet.
Open to both short, well-scoped tasks and longer-term engagements.