Vidhi Upadhyay

About

Independently built a production-grade, containerized enterprise agentic AI platform (stateful orchestration + RAG) focused on real-world reliability—guardrails, citation-based outputs, reranking, query rewriting, and evaluation harnesses to reduce hallucinations. Hands-on with OpenAI SDK, CrewAI, and LangGraph, and has delivered AI solutions for non-technical NGO stakeholders via demos and practical POCs.

Experience

Education

Key Strengths

• Built and deployed a production-style enterprise agentic AI platform from scratch
• Reliability-first approach to reducing hallucinations via guardrails, validation, and citations
• Strong RAG engineering: semantic chunking, metadata-aware retrieval, reranking, query rewriting
• Designs modular, observable agent workflows with measurable evaluation metrics
• Creates automated evaluation harnesses and benchmarks; uses monitoring/logs/traces to prevent regressions
• Able to translate non-technical stakeholder needs into technical AI use cases (NGO project) and deliver demos/POCs
• Took an internal GenAI knowledge assistant from RAG prototype to production used by multiple teams
• Strong hallucination-mitigation via improved retrieval (chunking, metadata filtering, hybrid search) and prompt guardrails
• Production-readiness mindset: evaluation datasets plus logging/tracing to monitor and prevent regressions
• Latency/accuracy optimization focused on retrieval efficiency (Top-K tuning, caching, precomputed embeddings, efficient vector indexes)
• Real-time LLM workflow debugging using end-to-end traces and query replay to isolate root causes
• Pragmatic incident mitigation with guardrails/fallbacks to maintain service continuity
• Effective developer-facing demos/workshops using interactive, hands-on live testing rather than slideware
• Cross-functional partnership with sales: translating business pain points into technical use cases and building customer-specific POCs to improve late-stage conversion