MK Mayank Kumar

Amazon Web Services · Professional Services

Mayank Kumar

Senior Architect — .NET & Mainframe Modernization, Agentic AI on AWS

Tech debt starts accumulating the moment you write the code. Nineteen years in, I make a living undoing it: .NET off Windows and onto Linux, COBOL off the mainframe and into .NET, monoliths into services. AI has changed the economics of that work. It has not changed the part where someone still has to be accountable for what ships.

01

Summary

I am a Senior Architect at AWS Professional Services. Before that, nineteen years of C# and .NET, most of it with my hands on the keyboard. For the last four years the work has been application modernization with generative AI: .NET off Windows and onto Linux containers, and COBOL mainframe systems turned into .NET services using AWS Transform, Amazon Bedrock, Strands Agents and Kiro.

I also build and run agentic systems in production on Amazon Bedrock AgentCore: runtime, gateway, memory, identity, observability. A demo agent and a production agent are not the same thing. The difference is guardrails, evaluation, and being able to answer what the agent actually did when someone asks you six weeks later. Long background in healthcare (HL7, FHIR), banking and insurance, and years of leading distributed teams of 10+ engineers.

I write for the .NET on AWS Blog, review AWS drafts, and post about this work on my blog and on LinkedIn. I write those myself. Certified across AWS, Azure and Google Cloud (Solutions Architect Professional, ML Specialty, Azure Solutions Architect Expert, Professional Cloud Architect). AWS is where I actually work.

02

What I focus on

Agentic AI

Agentic AI on Amazon Bedrock

  • Multi-agent systems on Amazon Bedrock AgentCore: runtime, gateway, memory, identity, observability.
  • Strands Agents SDK for tool-using agents. MCP servers to expose enterprise systems without opening them up.
  • Model choice, prompt and context engineering, evaluation harnesses, a human in the loop where it matters.
Modernization

Mainframe & COBOL to .NET

  • Agentic pipelines with AWS Transform and Bedrock: analyze, decompose, generate, test.
  • Business rules pulled out of COBOL nobody has read in thirty years, and written down.
  • Target architectures on ECS/EKS and Lambda. Data off VSAM and DB2.
.NET on AWS

.NET Replatforming & Optimization

  • .NET Framework on EC2 Windows → .NET 8/9 on Linux containers. License cost out, release speed in.
  • .NET Upgrade Assistant and Microservices Extractor to assess and break up monoliths.
  • .NET Aspire, OpenTelemetry/ADOT and CDK, so the thing is observable and reproducible.
Delivery

Spec-Driven Engineering with Kiro

  • Kiro specs, steering and hooks, so AI-assisted development is repeatable and not a party trick.
  • CI/CD, IaC and a real test strategy. Generated code passes the same gates as hand-written code.
  • Enablement: the team should be able to run this after I leave. Otherwise I have not finished.
03

Experience

Senior Architect · Amazon Web Services (ProServe)

Sep 2022 – Present

Application modernization & generative AI · global enterprise customers

  • I lead .NET modernization engagements from assessment through production cutover. .NET Framework comes off EC2 Windows, .NET on Linux containers go in, the license bill drops and releases stop being an event.
  • Ran a COBOL → .NET mainframe modernization program with AWS Transform, Amazon Bedrock, Strands Agents and Kiro. Agentic pipelines read the legacy code, recovered the business rules, generated .NET services, and wrote the tests and the documentation nobody had written the first time.
  • Architect and operate agentic AI on Amazon Bedrock AgentCore: runtime isolation, tool access through the gateway, session memory, identity, and observability across the whole multi-agent workflow. The observability is the part people skip and then regret.
  • Set up spec-driven development with Kiro (specs, steering, hooks) so generated code follows the customer's standards and goes through the same review and tests as anything a human typed.
  • Used .NET Upgrade Assistant and Microservices Extractor to assess large monoliths and find the service boundaries that were already there.
  • Built observability into modernized systems with OpenTelemetry / ADOT, custom processors included, and put the infrastructure in AWS CDK with pipelines around it.
  • Deliver hands-on AWS workshops, write for the .NET on AWS Blog, review other people's drafts, and mentor in the ProServe .NET and generative AI communities.

Technologies Amazon BedrockBedrock AgentCore AWS TransformStrands Agents Kiro.NET 8/9 ECS / EKSLambda Step FunctionsAWS CDK OpenTelemetry / ADOTCOBOL Amazon Q Developer

Architect · Hitachi Vantara

Oct 2017 – Sep 2022

Architecture across healthcare, industrial and retail programs

  • UK healthcare organization (2019–2021): architect on one of the UK's largest healthcare initiatives, on site in the UK for three months. Designed every component of a national-scale platform moving clinical data over HL7 and FHIR. Getting that wrong is not a rollback.
  • US paints company (2021–2022): built an MLOps framework so models actually reached production, with the CI/CD and infrastructure automation to keep them there.
  • Security systems company (2019): started the program and led 10+ engineers across locations. Designed a containerized event-driven system on Docker Swarm, RabbitMQ, MongoDB and NiFi.
  • Grocery chain IoT platform (2018): architected the ingestion platform and the hardware integration, with teams in Bangkok, Pune and Hyderabad.
  • Digital GDC initiative (2017–2018): built the technical backbone from nothing, led 10 engineers, owned architecture review, code review and the build and deploy pipeline.

Technologies .NET CoreReactJSAngular HL7 / FHIRDocker SwarmRabbitMQ MongoDBPython Azure (App Services, Functions, Event Hubs, Azure ML, AD B2C, DevOps) Google Cloud IoT

Architect · ValueMomentum Software Services

Nov 2015 – Oct 2017

Insurance · conversational systems and container platforms

  • Architected a claims and policy chatbot on Microsoft Bot Framework, back when a chatbot was a state machine and you had to earn every answer. Owned the authentication model and how policyholder data was handled.
  • Led a Docker Swarm proof of concept for the client's DevOps pipeline and worked with the product team and stakeholders on the platform decision that came out of it.

Architect / Lead · Microsoft Corporation India

Jan 2011 – Nov 2015

Enterprise solutions for retail, energy, IT services and legal clients

  • Retail giant: architect and dev lead for a team of 5, rebuilding organization-wide meeting controls after a Lotus Notes to Office 365 migration.
  • Oil giant: architect and onsite coordinator for a Word-based "Well File" solution that replaced physical document folders. The users never had to leave Office, which is why they used it.
  • IT services company: designed dynamic document classification and Information Rights Management driven by a web service. Owned its security and how far it scaled.
  • Legal industry product: shipped two releases of a distributed legal-document search product, and worked out how to intercept drag-and-drop on Office documents through Windows native APIs. Nobody had done that one before.

Developer · Sapient Consulting & Cognizant Technology Solutions

Nov 2007 – Apr 2011

Where I learned: WinForms, WCF and high-volume data

  • Kept a large WinForms and WCF trading and operations system running for an oil major (Sapient).
  • Built non-blocking UI for a banking application so people could keep working while large data sets loaded behind them (Cognizant).
04

Technical skills

Generative & Agentic AI

  • Amazon Bedrock
  • Bedrock AgentCore
  • AWS Transform
  • Strands Agents SDK
  • Kiro
  • Amazon Q Developer
  • Model Context Protocol (MCP)
  • RAG
  • Prompt & context engineering
  • Agent evaluation & guardrails
  • Amazon SageMaker

AWS Platform

  • EC2
  • ECS
  • EKS
  • Lambda
  • Step Functions
  • API Gateway
  • S3
  • DynamoDB
  • RDS / Aurora
  • CloudFront
  • IAM
  • AWS CDK
  • CloudFormation
  • CloudWatch
  • ADOT
  • AWS Glue
  • Well-Architected

.NET & Modernization

  • C#
  • .NET 8 / 9
  • ASP.NET Core
  • .NET Aspire
  • Minimal APIs
  • .NET Upgrade Assistant
  • Microservices Extractor
  • COBOL → .NET
  • WCF → REST/gRPC
  • Entity Framework Core
  • TDD
  • C / C++
  • Python

Architecture & DevOps

  • Microservices
  • Event-driven architecture
  • Domain-driven design
  • Infrastructure as Code
  • CI/CT/CD
  • Docker
  • Kubernetes
  • OpenTelemetry
  • Design patterns
  • HLD / LLD
  • TLS/SSL
  • Agile

Domains

  • Healthcare (HL7, FHIR)
  • Mainframe modernization
  • Banking & financial services
  • Insurance
  • Retail & IoT

Also worked with, and have not forgotten

  • Azure App Service / Functions
  • Azure ML
  • Azure DevOps
  • Azure AD B2C
  • ARM templates
  • Google Cloud (IoT, Pub/Sub)
05

Writing & speaking

.NET on AWS Blog

Building AI-Ready Applications with Amazon Bedrock and .NET Aspire

Wiring foundation models into a distributed .NET application you can actually observe and ship.

Read → .NET on AWS Blog

Developing Custom Processors using OpenTelemetry in .NET 8

Shape telemetry at the source. You get cleaner data and a smaller bill.

Read → LinkedIn article

Use Amazon Q Developer to query Splunk logs

Pointing a coding agent at your logs instead of learning yet another query language.

Read → blog.mayanks.page

Personal blog: generative AI, .NET, cloud engineering

Longer notes. RAG over chat archives, Glue unit tests in pipelines, telemetry correlation, and whatever else I broke that week.

Browse → LinkedIn

Posts, and the occasional argument

AWS Transform, Kiro, AgentCore, and a running commentary on which parts of the AI boom are real. I write them myself.

Follow →
06

Education & certifications

Education

  • B.E. Electronics — Bharati Vidyapeeth College of Engineering, Pune

AWS certifications

  • AWS Certified Solutions Architect – Professional
  • AWS Certified Machine Learning – Specialty
  • AWS Partner: Generative AI Essentials

Other cloud certifications

  • Microsoft Certified: Azure Solutions Architect Expert (AZ-303, AZ-304)
  • Google Certified Professional Cloud Architect
Earlier Microsoft certifications
  • Exam 70-532 — Developing Microsoft Azure Solutions
  • Exam 70-534 — Architecting Microsoft Azure Solutions
  • Exam 70-573 — SharePoint 2010 Application Development
  • Exam 70-511 — Windows Applications Development, .NET 4
  • Exam 70-536 — .NET Framework 2.0 Application Development Foundation
  • Exam 70-513 — WCF Development, .NET 4

Need this as a document?

Take a clean single-column copy and edit it in Word, Google Docs or Pages. No sign-up, no email required.