Skip to content

About Swarnava Dutta, Lead AI Architect

Lead AI Architect in India, with 6+ years across data engineering and enterprise AI delivery. I work across retrieval, agent orchestration, evaluation, deployment, and team handoff.

About me

I build AI systems teams can operate after launch.

I'm an AI engineer and architect based in India. For 6+ years I've worked across data engineering, search, and generative AI, where answer quality, latency, operating cost, and compliance all affect the design.

Multi-Agent SystemsFine-tuning & LoRARAG ArchitectureKnowledge Graphs and GraphRAGMCP WorkflowsEvaluation Pipelines

What I focus on

My work combines agent orchestration, retrieval, evaluation, and deployment. Years in cloud data platforms and automation shaped how I handle data quality, failure recovery, and operating ownership.

Much of my recent work centers on LangGraph orchestration, RAG, and domain-specific AI products. Recent platforms have supported 10K+ daily queries while maintaining 99.9% uptime in live environments.

The outcomes are concrete: lower latency, fewer manual steps, and faster decisions. The systems represented here delivered efficiency gains as high as 67% and were handed over with monitoring and documentation.

How I work

The standards I use when building AI systems.

Quality

Measure the behavior before you trust the behavior.

I establish a baseline, task-specific checks, and regression tests before increasing traffic or scope.

Test retrieval quality before scaling traffic.

Review answers against task-specific criteria.

Tie architecture decisions to user and business measures.

Reliability

Design for production traffic and messy data.

Production systems need bounded workflows, fallbacks, and enough telemetry to diagnose failures quickly.

Keep agent roles and handoffs explicit.

Set latency, fallback, and caching rules early.

Plan monitoring and team ownership with model behavior.

Usefulness

Remove work people should not have to repeat.

A useful AI system shortens a workflow, improves a decision, or removes repetitive steps people can measure.

Keep the success measure visible from discovery through launch.

Add security, guardrails, and handoff plans early.

Leave the system documented and practical to extend.

Keep exploring

SkillsServicesProjectsExperienceCertificationsFAQBlogContact