Mukul Varshney

GenAI engineer building production multi-agent RAG systems on a 20-year foundation of enterprise architecture and delivery. Currently working on ReadyTech's Planner Assist AI project — an agentic assessment engine for Australian local-government planning — covering agent orchestration (LangGraph), hybrid retrieval, multi-LLM routing with cost observability, and guardrail/validation layers. Claude Certified Architect – Foundation (CCA-F) and Azure AI Engineer certified. Background as certified OutSystems Architecture Specialist leading onshore/offshore teams, with earlier .NET full-stack product engineering. CORE AI SKILLS • Agent orchestration: LangGraph StateGraph (parallel fan-out/fan-in, conditional routing, sequential reflection pipelines, incremental regeneration), LangChain — TypeScript & Python • RAG & retrieval: hybrid search (semantic + BM25 + Cohere reranking), PostgreSQL/pgvector & Qdrant, contextual + multimodal embeddings, LlamaParse ingestion, retrieval evaluation (hit@k, MRR, ablation studies) • Quality, safety & evals: input guardrails (PII redaction via Presidio, prompt-injection detection), deterministic grounded-citation validators, golden/eval-set design, LLM-as-judge, Langfuse observability & per-node cost tracking • Multi-LLM engineering: Claude (AWS Bedrock), GPT, Gemini, Cohere; structured outputs with Zod schema enforcement; prompt-caching-aware prompt design; factory-pattern instantiation for queued pipelines • MCP (Model Context Protocol): designed and shipped 3 servers — tools, resources, sampling, roots, progress notifications; stdio + Streamable HTTP transports; DNS-rebinding protection • Platform & stack: TypeScript (Fastify, Drizzle, BullMQ/Redis), Python, PostgreSQL, Docker, GitHub Actions CI/CD, ArgoCD, AWS (Bedrock, ECR, S3) + Azure (Blob, ACR)