Naini Vivekanand

I build agentic AI systems that go beyond notebooks and actually run reliably in production. I’m focused on LLM orchestration, multi-provider routing (including Claude/Anthropic) with health-scored failover, and making tools/agents safe with real-time guardrails and human-in-the-loop approvals.

I also design and evaluate retrieval-augmented generation pipelines—combining keyword and semantic search with reranking—and I obsess over test quality. Across my projects, I’ve built automated evaluation harnesses and hundreds of checks to catch regressions, hallucinations, and edge-case failures, so the systems behave predictably when real users and messy inputs show up.