Vivek Sahu

I’m a software engineer with strong CS fundamentals and a bias toward correctness. I ship and test production systems end-to-end, and I focus on deterministic, auditable logic over unnecessary complexity—especially when decisions can’t be “mostly right.”

In my current campus leadership role, I design and deliver hands-on AI/ML workflows for 50+ learners, including model fine-tuning, vector database integration, and RAG evaluation. I also raise code-quality across a 100+ member developer community by mentoring on PR review and Git-based collaboration norms.

My projects reflect how I build: MandateCheck enforced payment authorization with a zero-LLM decision path using a fully rules-based mandate-verification engine, eliminating non-deterministic failure modes. I’ve also shipped live systems with real-time WebSocket event feeds and operational control (a manual kill switch) for safe demonstrations.

On the AI side, I build RAG and multi-agent pipelines that prioritize reliable outputs and security. For example, Candor uses adversarial Advocate/Challenger agents with an Arbitrator to verify tool outputs before streaming results, while Agent Rail improves latency and reliability with bounded concurrency, rate limiting, and CI hardening.