Skip to main content
Aditya Tripathi

Aditya Tripathi

AI engineer who took a multi-agent agentic system from prototype to production for 2500+ healthcare users

Talent
DelhiMember since September 21, 2026
PythonFastAPIOOPstreamingLangGraphconditional edgesmemoryLangChainstructured outputsgrounded abstentionprompt injection defenceHybrid retrievalpgvectorHNSWrerankingmetadata filteringRAGASLangfusefailure dashboardsproduction samplingNode.jstRPCREST APIsVercelPostgreSQLRedisNext.js 15
·

Hire this person

Send a job offer directly to this candidate

PythonFastAPIOOPstreamingLangGraph

About

AI engineer who took a multi-agent agentic system from prototype to production for 2500+ healthcare users — LangGraph-orchestrated planner-executor-verifier architecture with hybrid RAG, eval harnesses gating CI, and Langfuse observability. Eval improved answer quality 55% → 82%, cut wrong responses 9% → 2%.

Experience

Software Engineer — AI & Backend at DocNow (2024-12 – Present)

  • Designed and deployed a production agentic system using LangGraph — planner node classifies intent via structured outputs, retrieval node runs hybrid BM25 + pgvector search with reranking over 1100+ tests/packages, action node calls booking APIs via tool calling, verifier node enforces grounded abstention and escalates to a human, eliminating hallucinated booking confirmations.
  • Engineered state management, memory, and dynamic loop control — Redis-backed memory persistence, per-turn context trimming, conditional edges for retry-on-low-confidence, and a hard 5-step loop cap to bound latency and cost.
  • Built a comprehensive eval framework with an 80-query golden set — RAGAS-style automated scoring (faithfulness, context precision, answer relevance, hallucination rate) gated in CI on every deploy; improved context precision 55% → 82%, cut confident-wrong responses 9% → 2% through chunking strategy and reranker tuning.
  • Established production observability with Langfuse — span annotations per agent node, failure dashboards, prompt injection defence, guardrails on tool outputs, 10% production sampling for drift detection; diagnosable from trace data alone.
  • Delivered 147 production APIs (Python + FastAPI microservice, Node.js, PostgreSQL, 26-table schema) serving 2500+ patients across 50+ cities; 99.99% webhook reliability via idempotent processing, retry queues, dead-letter recovery, and lease-based DB locking.

Full-Stack Web Developer Intern at Badaya Fitness Sports Pvt Ltd (2024-06 – 2024-12)

  • Migrated to Next.js App Router; built secure, query-optimized REST APIs with robust async error handling over complex nested PostgreSQL data.

Education

B.Tech in Electrical Engineering – Bundelkhand Institute of Engineering and Technology Jhansi (2022 – 2026)

Reviews

Related