existing to remainnew work
Solution Architect at Tata Consultancy Services, fifteen years in one line of work: taking enterprise estates that cannot go down - mainframe, COBOL, legacy .NET - and rebuilding them as Java and .NET microservices on Azure. Lately with GenAI reading the code that no one left to explain it.
← 2011 ‐ 2026 · overall →
Every modernization starts the same way: a single artifact that does everything, that nobody can deploy on a Tuesday. Split it and you buy independence - and pay for it in network calls. Here's the trade, live.
The honest half: splitting adds network latency, eventual consistency, distributed tracing and per-service operations. Cut along business capabilities - bounded contexts - not technical layers, and only where the coupling is actually costing you. A well-run monolith beats a badly-drawn mesh every time.
The hard part of a mainframe migration was never syntax - it's that the business rules live in code no one has read in twenty years. An agentic pipeline reads it, maps it to a context, rewrites it, then argues with itself about the edge cases.
IDENTIFICATION DIVISION.
PROGRAM-ID. CALC-INTEREST.
DATA DIVISION.
WORKING-STORAGE SECTION.
01 WS-PRINCIPAL PIC 9(9)V99.
01 WS-RATE PIC 9(2)V9(4).
01 WS-TERM PIC 9(3).
01 WS-INTEREST PIC 9(9)V99.
PROCEDURE DIVISION.
COMPUTE WS-INTEREST ROUNDED =
WS-PRINCIPAL * WS-RATE
* WS-TERM / 1200.
DISPLAY 'INTEREST: ' WS-INTEREST.
STOP RUN.
// run the pipeline to generate
Four spans, one system under continuous migration. Open any span for what actually shipped.
COBOL estates carrying decades of business rules, with no surviving documentation and no author left to ask.
Architected a GenAI-driven migration that builds contextual understanding of the legacy system first, then generates Java - reviewed and test-gated rather than trusted.
Recognized internally as Contextual Master - GenAI for Mainframe Modernization (2024).
Migration handled as a series of one-off projects, with quality depending on which team drew the short straw.
Led lifecycle tooling - Cloud Console, Migration Factory Hub, Well-Architected reviews - so assessment and landing zones became repeatable.
A standard path from legacy estate to cloud-native, reusable across engagements.
Compliance checking across networks and operating systems done manually, differently, by every team.
Took it from research to product: policy definition, a compliance engine, and integrations for mixed enterprise infrastructure.
Two granted patents - US (2017) and India (2024) - plus a shipped service.
Collaborative mind-mapping tools want your data on their servers and a dependency tree to match.
Zero-runtime-dependency vanilla JS, self-hostable in one command; real-time collaboration on Cloudflare Workers and Durable Objects with GitHub OAuth and JWT role-based access.
MIT-licensed, maps stored in the user's own repo. Source →
The other half
PhD scholar at IIIT Nagpur, published in IEEE Transactions on Instrumentation & Measurement (2026) - the institute's first quantum-computing paper. Different site, different demos: a live qubit, Grover search, and the Simon → adapt → Grover pipeline.
Architecture reviews, modernization strategy, and the awkward question of whether you should split that service at all.