AI & GenAI Solutions Architect · 16+ years · Financial services

Taking enterprise AI from concept to governed production.

I architect LLM, RAG and agentic platforms with the security, compliance and reliability that regulated businesses demand — grounded in sixteen years of delivering enterprise systems for Citibank APAC, Citibank India and USAA.

Start a conversation View LinkedIn
Portrait of Karthik Sukumaran
Caltech CTMEPost Graduate Program in AI & Machine Learning
Experience with Citibank HCLTech USAA Polaris FT
3 days → 1 hour
Design-to-code effort, with the agentic AIForce D2C platform I designed and built
9 countries
Citibank APAC content platform owned end to end, development through production
Zero
SAST and DAST findings after re-architecting the legacy codebase into microservices
3 weeks → 30 min
Hyperledger Fabric environment provisioning, via the CoTrust Blockchain Console

How I help

Architecture that leadership can trust — and engineers can build.

01

Take GenAI into production

Reference architectures for LLM, RAG and agentic platforms — multi-model orchestration across OpenAI and Claude, vector retrieval, evaluation and feedback loops — so pilots become products.

02

Modernise core platforms safely

Re-architect legacy systems into microservices while the business keeps running — shared business layers, secure-coding standards, and upgrades coordinated across countries and teams.

03

Build governance in, not on

AI usage governance, guardrails, disaster-recovery drills and compliance-mandated change windows designed into the architecture from day one — the way regulated industries need it.

Selected case studies

Outcomes, not just architecture diagrams.

GenAI · Agentic

HCLTech AIForce D2C

Designed & built · HCLTech · 2021 – present

3 days → 1 hour average design-to-code effort

Challenge

Turning Figma designs into production-ready front-end and CMS code was slow, manual work — around three days per design on average.

Approach

An agentic design-to-code platform: multi-model LLM orchestration (OpenAI, Claude), RAG over ChromaDB and Pinecone, and autonomous agents that interpret UI, generate code and refine it through feedback loops. Containerised on Azure, with a companion plugin published to the Figma Community.

Result

Production-ready React, Angular, Next.js and CMS output in about an hour, held to enterprise standards for performance, accessibility and maintainability.

Modernisation · Banking

Citibank APAC content platform

Manager · Citibank N.A. · 2018 – 2021

9 countries · zero SAST/DAST findings

Challenge

A legacy SOAP-based content management platform serving Australia, China, Hong Kong, Singapore, Thailand, Malaysia, Vietnam, the Philippines and Indonesia, in a highly regulated environment.

Approach

Led the re-architecture into microservices with one shared business layer serving both legacy SOAP consumers and new REST APIs; embedded secure-coding standards and code reviews; upgraded the full stack (RHEL, WAS, IHS, database, Documentum, Captiva) in every market; ran cross-team DR drills.

Result

A codebase with zero SAST and DAST findings, a platform upgraded across all nine markets, and maintenance delivered within the change windows set by business and compliance.

Blockchain

CoTrust Blockchain Console

Centralised Hyperledger Fabric setup and administration, replacing about three weeks of error-prone manual provisioning.

3 weeks → 30 minutes

Scale

High-volume document migration

A multithreaded migration job that moved an enterprise document estate between storage platforms.

3.5 million documents in 8 hours

My approach to enterprise AI

Four principles behind every platform I design.

Shaped by years in banking, where reliability, auditability and security are non-negotiable.

  1. Grounded in your data

    RAG over governed enterprise sources, so answers are anchored in the organisation's own context.

  2. Model-agnostic

    Multi-model orchestration and routing across providers such as OpenAI and Claude — no single-vendor lock-in.

  3. Agentic where it pays

    Autonomous agents with iterative feedback loops (LangGraph, MCP) for work that benefits from planning and self-correction.

  4. Governed by default

    Usage governance, guardrails and evaluation designed in, with the observability operations teams expect.

Experience

Sixteen years, four roles, one thread: enterprise systems that work.

  1. Sep 2021 — Present

    Senior Technical Architect — GenAI Solutions

    HCLTech

    Architect of AI-driven engineering platforms; owns architecture, design patterns and AI governance from concept to production. Leads GenAI and blockchain solution design for enterprise clients.

  2. Aug 2018 — Sep 2021

    Manager

    Citibank N.A.

    Strategic and technical leader for the content management platform serving nine APAC markets. Proxy Product Owner; owned upgrades, disaster recovery, compliance and team development.

  3. Mar 2013 — Aug 2018

    Senior Technical Lead

    HCLTech

    Led Agile delivery of enterprise content management projects, including onsite client engagements; Blockchain Architect and core member of the HCL CoTrust platform (Hyperledger Fabric, Ethereum).

  4. Dec 2009 — Mar 2013

    Associate Consultant

    Polaris FT

    Delivered a Documentum 6.5 document tracking and data-entry workflow for Citi Wealth Advisory, Citibank India, including a single account-opening form across banking, card and loan products.

Expertise

Depth across the stack.

GenAI & agentic AI

LLM orchestration · OpenAI · Claude · RAG · LangGraph · MCP · Prompt engineering · Pinecone · ChromaDB · Guardrails · LLM evaluation

Architecture & integration

Microservices · Event-driven systems · REST & SOAP · Blockchain (Hyperledger Fabric, Ethereum, Corda) · Documentum & Captiva

Cloud, data & engineering

Azure · AWS · IBM Cloud · Docker · Kubernetes · Python · Node.js · Java · MongoDB · Oracle · Elasticsearch

Delivery & governance

IT roadmaps · Agile & Waterfall · ITIL · Secure coding (SAST/DAST) · Disaster recovery · Distributed team leadership

Credentials & recognition

Education, certifications and awards.

Education

  • Post Graduate Program in AI & Machine LearningCaltech CTME
  • B.E., Electronics & Communication EngineeringAnand Institute of Higher Technology, Anna University

Recognition

  • O∞O-Infinity Award, 2021–2025Outstanding rating four years running, HCLTech
  • O∞O-Infinity Award, 2013–2017Outstanding rating four years running, HCLTech
  • GEMGone Extra Mile Award, 2011Citibank Auto Fulfillment Unit project, Polaris

Certifications

  • ITIL Foundation
  • AWS Cloud Practitioner
  • MongoDB SI Associate
  • Corda Certified Developer
  • Corda for Business Professional
  • New Product Development — IIM Bangalore
  • Blockchain for Business — Linux Foundation
  • IBM Blockchain Essentials & Developer Foundation

Writing

Architecture notes.

All posts →
Distributed transactions

Saga

How to keep a multi-service business operation consistent with a chain of local transactions and compensating actions — orchestrated or choreographed.

Read → Reliable messaging

Transactional Outbox

How to save business data and the event that describes it in one database transaction, then relay it to Kafka or a queue — solving the dual-write problem with polling or CDC.

Read → Resilience

Circuit Breaker

How a circuit breaker detects a failing dependency, stops calling it, and probes for recovery — closed, open, and half-open states, threshold tuning, Resilience4j and Polly examples, and mesh-level outlier detection.

Read →

Planning an AI platform or a modernisation programme?

I'm glad to compare notes on GenAI architecture, agentic platforms and AI adoption in regulated industries.

karthik.sukumaran@gmail.com Connect on LinkedIn