Nine projects
Government platforms, enterprise delivery and AI. Four have a short case study with the details one click away. Five are short summaries.
Projects
8th Pay Commission demand portal
Replaced an Excel sheet with a working portal in about a month, for the Commission that reviews pay for about 70 lakh people.
Green Hydrogen certification portal
India's national certification portal for green hydrogen, launched by the Minister on 1 July 2026. No process existed, so it was defined first.
Wind-energy application portfolio
A loss-making unit supporting about 55 applications for wind turbines, turned to about 15% profit in six months.
National cooperative registry
Moved a paper-based national registry online. The law had to be amended first, then the platform was built.
Ranking framework for government websites
Built with IIT Delhi and IIM Indore, covering 150+ parameters. NeSDA uses it today.
The short version
The problem
NeSDA is the national framework that assesses and ranks government websites. The departments it assessed did not report to the project team.
What was done
Used the weight of senior government offices instead of relying on persuasion: Secretary-level communication, CERT-In direction and a named contact in each organisation. The automated scoring was checked against human auditors before anyone relied on it.
The result
Around 100 sites were assessed with 20 auditors, and the scores were tuned until they sat within ±10% of the auditors. India placed among the top five G20 economies in the benchmark, and the framework continues as AKAM (Aanklan).
Indian visa system, IVFRT 2.0
Visa, e-visa, OCI and NRI services consolidated onto one national platform.
The short version
The problem
Old, separate and slow systems with data-sync problems, serving foreign nationals and tourists at the border, with practice that differed from port to port.
What was done
Designed from watching officers and travellers work, not from the written specification. Kept the separate databases for the first phase, because a port has no maintenance window and merging them carried the most risk.
The result
One consolidated platform in staged rollout: a two-seaport pilot, seaports live, airports piloting and land ports to follow.
National Portal of India
One front door to 20+ government sources, in 22 languages.
The short version
The problem
A static directory of disconnected portals. The aim was to bring reliable information from many government sources into one place, in 22 languages.
What was done
Treated translation as something to measure. People reviewed the machine output, accuracy was checked against agreed language criteria, and volunteers did blind surveys. A custom content system brought the different sources together.
The result
About 80% of machine output was accepted as translated, from an early error rate near one sentence in ten. Search and summaries were planned for the next phase.
Skillchain
A blockchain credential platform taken to a unit-wide EY pilot.
The short version
The problem
Credentials were issued as NFTs on Quorum for an EY unit. Joining was voluntary, so the product had to earn its own use.
What was done
Gathered requirements from all four user groups before building, and cut the NFT marketplace to protect the core.
The result
Around 80% of people joined voluntarily. The capability continued through 2024 and now lives in EY Skills and EY Badges on Degreed.
Mainframe automation and blockchain
Reconciliation automation for a UK bank, saving about £200,000 a year.
The short version
The problem
A manual card-dispute reconciliation on an old mainframe. The real blocker was a permission that had never been set up.
What was done
Got the permission sorted through the client instead of working around it, and rebuilt the pipeline instead of scripting over the manual steps.
The result
About £200,000 a year saved by the client's own calculation, and 40% fewer reported defects at six months. Two blockchain proofs of concept reached production, and the first blockchain lab on TCS's own infrastructure was set up.
Where AI is part of the work
Some of this is live, some is approved and in procurement, and some is planned. The status column says which.
| Project | What the AI does | Status | How it is checked |
|---|---|---|---|
| Green Hydrogen portal, Phase 2 | AI to check submissions, spot unusual emissions data and support verifiers | Approved, in procurement | Compared with cases people have already answered by hand |
| 8th Pay Commission portal | AI to spot duplicate demands and find the evidence a demand refers to | Planned, next stage | Acceptance rules agreed before anything is built |
| Wind-energy portfolio | AI to route and close support tickets | In production | Tickets closed and routed correctly, with no false positives found |
| National Portal of India | Machine translation into 22 languages, with human review | Live | Accuracy against agreed language checks, plus blind surveys with volunteers |
| NeSDA website scoring | Automated scoring of government websites | In use | Within ±10% of human auditors, across about 100 sites and 20 auditors |
Wherever AI helps with a decision in government work, AI advises and a named officer approves every important action. Every claim it makes must point to the document behind it.
I use Claude for documents, test cases and analysis on my projects. A person checks every output, and developers own every line of merged code.