Headline

Staff AI/ML Engineer · RAG Pipelines · Agentic AI · LLM · Vector Databases · MCP | Building Production AI @ Intuit | Ex-Visa · Ex-TCS | AI Educator

About

Get in touch with me: https://topmate.io/sumannath/ I've spent 14 years making complex systems reliable. The last 3, I've been doing that for AI. That's a harder problem than most people admit. Building a RAG pipeline that demos well takes a weekend. Building one that handles 10,000 concurrent tenants, respects data boundaries, invalidates stale knowledge automatically, and explains its own failures in real-time — that takes a fundamentally different approach. That's what I work on. --- At Intuit, I architect agentic AI solutions that bridge the gap between generative AI's promise and production-grade reliability. My work spans three areas: → AI Trust & Evaluation: Defining rigorous evaluation frameworks and real-time observability strategies so AI responses are accurate, safe, and auditable — not just impressive in a demo. → Agentic Systems: Designing multi-agent architectures using MCP (Model Context Protocol) that connect AI reasoning to live enterprise data — Splunk, ServiceNow, Confluence — with the reliability you'd expect from any mission-critical system. → Knowledge Engineering at Scale: Architecting high-throughput document ingestion pipelines using context-aware chunking, hybrid vector + graph retrieval (Milvus + Neo4j), and chunk-level cache invalidation for precise, secure knowledge utilisation. --- Before Intuit, at Visa, I built RAG pipelines for financial workflows, developed an MCP-based triage agent connecting to internal systems in real time, and led distributed systems architecture across high-volume payment platforms. And before Visa, 10+ years at TCS gave me something no architecture diagram teaches: how to lead teams of 50+ across India and the Netherlands, stay calm when a production system goes down at 2am, and make technology actually serve the business. --- Outside work, I love teaching AI/ML engineering and system design, helping mid-to-senior engineers break into top-tier AI roles. I also write about applied AI, RAG systems, and the real cost of building reliable intelligent systems at nathlabs.com. --- If you're building production AI systems, evaluating RAG architectures, or want to talk about what "trustworthy AI" actually means in practice — I'm always happy to connect. 📍 Bengaluru | Open to Staff / Principal AI Engineering roles where the hard problem is making AI reliable, not just impressive.

Location

Bengaluru

AI stack

anthropic apis

claude code

cursor

gemini

github copilot

hugging face

Work experience

Staff Software Engineer at Intuit, Feb 2026 - Present · 8 mos, Bengaluru, Karnataka, India, Architecting trustworthy agentic AI systems that close the gap between generative AI's promise and what production actually demands. → AI Trust & Evaluation: Designed rigorous evaluation frameworks and real-time observability pipelines to measure agentic response accuracy, safety, and reliability at scale — replacing manual spot-checks with automated, metric-driven validation. → Automated Prompt Optimisation: Leveraged GEPA (Genetic-Pareto prompt evolution) to automatically optimise prompts against ground truth datasets of 400+ entries across 4–5 evaluation metrics — reducing prompt iteration cycles from days to hours (~70% time saving) when datasets are updated, replacing manual trial-and-error with LLM-driven reflective evolution. → Agentic Architecture: Built multi-agent systems using Model Context Protocol (MCP) that connect AI reasoning to live enterprise data sources — enabling autonomous triage, root-cause analysis, and resolution recommendations across complex user issue workflows. → Knowledge Engineering: Architecting high-throughput document ingestion pipelines with context-aware chunking and hybrid retrieval (vector + graph) to ensure proprietary knowledge is accurately and securely surfaced for high-precision queries. → Production Reliability: Defined AI success metrics and feedback loops that validate model behaviour against real-world outcomes — ensuring agentic responses remain accurate and auditable as systems evolve.

Staff Software Engineer at Visa, Aug 2022 - Feb 2026 · 3 yrs 7 mos, Bengaluru, Karnataka, India, Led AI/ML engineering and distributed systems architecture across Visa's high-volume financial platforms, with a focus on intelligent automation and production-grade reliability. RAG & Agentic AI: → Developed RAG pipelines integrating LLMs with internal knowledge bases using LangChain — enabling intelligent document summarisation and automated support workflows. → Built an MCP-based triage agent connecting to Splunk, Clickhouse, ServiceNow, and Confluence in real time — providing autonomous root-cause analysis and historical insights across live production incidents. Distributed Systems & Platform: → Architected Python-based microservices for critical financial transaction processing, improving scalability and resilience across high-volume payment platforms. → Led architecture reviews providing guidance on scalability patterns, fault tolerance, and distributed system design for backend platforms. → Deployed containerised applications using Kubernetes, Jenkins, GitHub Actions, and Helm charts for consistent, reliable production deployments. Automation & Data: → Automated complex batch workflows using Python, Control-M, and Bash — achieving an 80% reduction in manual effort and significantly improving data consistency. → Optimised database performance across MySQL, DB2, and Cassandra through query refactoring and schema optimisation — resulting in a 40% improvement in data access times. → Built scalable data processing pipelines with Pandas and Apache Spark for log ingestion, dataset enrichment, and analytical reporting across distributed systems.

Staff Engineer / Technical Lead at Tata Consultancy Services, Jan 2012 - Aug 2022 · 10 yrs 8 mos, Kolkata, West Bengal, India, Over a decade at TCS, I grew from individual contributor to Technical Lead — managing 50+ engineers across India and the Netherlands, delivering enterprise solutions for global clients across steel, manufacturing, and financial services domains. Technical Leadership: → Led Application Maintenance and Support (AMS) for 3+ years as primary escalation point for production outages, conflict resolution, and SLA management across mission-critical client systems. → Directed a diverse team of 50+ associates across Netherlands and India, ensuring seamless service delivery, client satisfaction, and continuous process improvement. → Served as LnD SPOC (Learning & Development Single Point of Contact) — mentoring engineers, driving upskilling initiatives, and maintaining team motivation through sustained delivery pressure. Engineering & Delivery: → Delivered multiple functional designs for legacy-to-cloud migration projects, contributing to successful client proposals and solution architecture across global accounts. → Developed software using MySQL, DB2, Python automations, and Java across object-oriented enterprise systems. → Integrated AWS services (EC2, RDS, S3) to implement cloud requirements for client workloads, supporting the transition from on-premise to cloud-native infrastructure. → Designed RESTful APIs for authentication, authorisation, and service orchestration supporting mission-critical operations. Client Delivery (UK & Netherlands): → Served as primary technical liaison for client-facing delivery across European engagements — managing requirements, stakeholder communication, and delivery timelines across time zones.

Education

Bachelor’s Degree, Computer Science, West Bengal University of Technology, Kolkata

High School, All Subjects, Don Bosco School

Public links

github: https://github.com/sumannath