Headline

Senior Analyst @ EY | Full Stack Engineer × GenAI Specialist | Azure + Copilot Certified | Don’t use AI to vibe code, write better code using AI while vibing

About

I'm a GenAI Engineer at EY designing and shipping production-grade, multi-agent LLM systems solving real enterprise problems. Over 3 years, I've architected agentic platforms transforming customer support, compliance, and enterprise automation workflows across retail, e-commerce, banking, and healthcare. What I build: - No-code AI platforms enabling non-technical teams to deploy context-aware voice agents, agentic workflows, and document processors without code - End-to-end observability stacks (Langfuse, Azure Application Insights, custom metrics), cuz black-box AI is a liability - Multi-agent systems with intelligent routing, human-in-the-loop escalation, PII safety, cost control via token budgeting, and Azure cloud-native deployment - Full-stack applications combining React/Next.js frontends with FastAPI/Node.js backends, vector RAG over PostgreSQL/Milvus, and secure authentication (Okta OIDC, Google SSO) Core expertise: - Agentic AI: LangGraph, LangChain, OpenAI Agents SDK, Microsoft Agent Framework, Model Context Protocol (MCP), Agent-to-Agent communication - Evaluation & Observability: LLM-as-a-judge, Langfuse, Prometheus, Grafana, sentiment analysis, structured logging - RAG & Context Engineering: Vector databases (pgvector, Milvus), semantic search, harness engineering, Pydantic schema validation, SQLModel - Cloud & DevOps: Azure Container Apps, Azure Functions, Durable Functions, Docker, microservices, CI/CD, Git - Databases: Cosmos DB, MongoDB, Firestore, PostgreSQL - Languages & Frameworks: Python (advanced), TypeScript, JavaScript, FastAPI, Flask, React.js, Next.js, Express.js Recent wins: Architected voice platform improving first-call issue resolution via intelligent agent routing and persistent multi-layered context memory Designed three-portal agentic workflow builder (admin/operator/end-user) with model-agnostic LLM support (OpenAI, Microsoft, fine-tuned models) Built observability & compliance platforms with real-time streaming (SSE), PII redaction, LLM output verification, and cost-aware session management At AlphaDnA: shipped asset visualization app (20% decision-speed, 50% API response improvement), full-stack product discovery with vector RAG, document translation platform with passwordless auth Certifications: AZ-900, AZ-204, AI-900, AI-103 (Azure AI Apps & Agents), GH-900, GH-300 (GitHub Copilot) Currently: Building production-grade agentic templates and open-sourcing knowledge (github.com/RTopdar) Exploring advanced RAG, LLM safety, fine-tuning strategies

Location

Greater Kolkata Area

Work experience

Senior Analyst at EY, Mar 2025 - Present · 1 yr 5 mos, Kolkata, West Bengal, India, Designing and deploying production-grade, multi-agent LLM systems transforming enterprise workflows. Leading three agentic platforms for retail, e-commerce, healthcare, and compliance. Voice Multi-Agent Platform: Built no-code system enabling teams to configure voice agents with distinct personalities and escalation logic. 40% improvement in first-call issue resolution via intelligent routing and persistent multi-layered memory. End-to-end observability using Langfuse and custom Azure metrics. Cost control via per-user token budgets and circuit breakers. PII redaction and LLM output verification by default. Twilio IVR for offline access. Deployed on Azure Container Apps with autoscaling. Tech: OpenAI Agents SDK, Langfuse, Temporal, Azure Functions. Agentic Workflow Builder: Multi-tenant platform for admins to design workflows visually, assign agents, connect tools, deploy without code. Model-agnostic engine supports OpenAI, Microsoft, fine-tuned models. MCP + A2A integration for external tool connections. Three-portal architecture (admin/operator/end-user). Smart orchestrator handoff logic reduces resolution path. In-house observability platform tracks per-agent performance in real-time. Secure isolated vector databases per department for RAG without cross-domain leakage. Microservice architecture with independent scaling and automated failover. Tech: Microsoft Agent Framework, LangGraph, MCP, Azure Container Apps. Document Processing Platform: Redesigned monolithic processor into microservices. MCP client architecture connects to compliance tooling servers. Azure Durable Functions orchestrates long-running pipelines. Real-time progress streaming via Server-Sent Events to clients. Reduced deployment time and scaling costs. Tech: Azure Durable Functions, MCP client/server, SSE.

Associate Web Developer at Gnosis Lab, Aug 2023 - Mar 2025 · 1 yr 8 mos, Kolkata, Develop front-end applications to visualize data provided by the client or back-end team

Full Stack Engineer at AlphaDnA Pvt Ltd, Jun 2024 - Feb 2025 · 9 mos, Kolkata, West Bengal, India, Developing high-performance React.js applications for data visualization with optimized loading times and superior user experiences. Optimize backend systems by developing efficient RESTful APIs and refining schemas, reducing response times and code complexity Deploy applications on Google Cloud Platform (App Engine, Cloud Run, Compute Engine), employing various deployment methods, version control, and CI/CD pipelines, achieving zero downtime for critical applications. Utilize Firebase storage and Firestore database to build secure and scalable file management systems Follow SRE best practices in deployment, enhancing system reliability. Collaborate with partners to manage project priorities effectively, resulting in improved project timelines and successful client outcomes.

Content Writer at Cannibals Media, Jan 2021 - Aug 2023 · 2 yrs 8 mos

Education

Bachelor of Technology - BTech, Computer Science, University of Engineering & Management (UEM)

Future Campus School - India