By Vikas Sawarakr

There is a very specific kind of tired that only HR people know.

It's 9:47 PM. The office is empty except for one laptop glowing in the dark, and one person copy-pasting a candidate's name into an offer letter for the eleventh time that day. Change the name. Change the CTC. Change the date of joining. Check the address didn't wrap weird. Check the designation matches what the client sent in the email. Save. Rename the file. Do it again for the contract letter. Do it again for the relieving letter, except this time remember there are two versions — one for "Mr." and one for "Ms." — and if you grab the wrong one, someone's experience certificate will misgender them on the way to their next job.

Now multiply that by 300 to 350 times a year, because that's roughly how many joiners, exits, and role changes MITS Staffing processes annually across its 600 to 700 person payroll. Two employee groups — BNP Paribas client staff and direct MITS2 hires — each with their own templates, their own quirks, their own ways of breaking if you're not paying attention.

Nobody got into HR to become a human find-and-replace function. And yet, for years, that's what the job quietly turned into.

So the team did something refreshingly simple: they stopped asking a person to do what a machine is actually good at, and let the human do what only a human can.

The Insight Nobody Says Out Loud

Here's the uncomfortable truth about most "HR paperwork": it isn't really HR work. It's data entry wearing an HR costume. The actual HR skill — reading a candidate, judging a salary structure, catching that a client's request doesn't smell right — that part was always safe from automation. It was the surrounding 80%, the transcription and formatting and triple-checking, that was eating people alive.

MITS built what they now call the AI Letter Writer, and the idea behind it is almost stubbornly simple: upload the source document — the offer email, the internal note, whatever raw information already exists — and let the system read it, understand it, and produce every letter a new joiner or exiting employee needs, correctly formatted, correctly branded, correctly calculated, in the time it takes to make a cup of chai.

Under the hood it's an ASP.NET Core application wired to GPT-4o through Microsoft's Semantic Kernel, reading Word documents with OpenXml and writing them back out the same way. But nobody in HR needs to know any of that. What they need to know is this: upload a file, wait a few seconds, download a zip folder with an offer letter, a contract letter, and whichever of the joining bonus, conveyance, relocation, or security-document letters actually apply — because the system reads the source text and figures out on its own which ones are needed. No checklist. No "did I forget the conveyance letter again."

The Part That Actually Matters

Here's what's easy to miss in a story about automation: this isn't a story about replacing anyone. MITS still has an HR team. What changed is what that team spends its evening doing. It used to be typing. Now it's reviewing — a five-minute glance at a generated letter instead of a forty-minute production of one.

And there's a quieter win buried in there too. Every generated letter is calculated the same way, every time, following the same salary logic, the same formatting rules, the same placeholder structure. The system doesn't get tired at 9:47 PM. It doesn't accidentally grab the wrong gender template because it's the eleventh letter of a long day. Consistency, it turns out, was the real gift — not speed.

Somewhere in MITS's office, there's a laptop that used to glow past midnight. These days it's dark by six. That's not a headline anyone puts on a product page. But if you've ever been the person still typing offer letters when everyone else has gone home, you know exactly why it matters.