March 2024, i got a remote job at a US based AI startup which was a flex back then, super grateful for them because despite of having no background in applied AI, they gave me the role. I was having a good time there (in fact i did some of the best work of mine that i still draw from) but somewhere i always wanted to come to BLR.

Jan 2025 was the time when i made the decision to switch, started applying to companies, gave bunch of interviews while i was wrapping up. After 2-3 weeks of extensive interviews, rejections, learnings, I came across a tweet about sarvam, started exploring about the company, found that they have an opening, and on 17 Feb 2025 I sent an email (find it below)

After around 10 days, i heard back, interviews happened, and exactly 1 month after the email, 17 March 2025, I was in indiranagar bengaluru, joining my first in-office stint at sarvam.

what my time actually looked like

what started as a frontend role, eventually started evolving into much more. it wasn’t one neat job the whole time. more like multiple roles that kept trading places as the company moved.

started in the voice agents — it wasn’t a product where people come once, do the stuff and go away. it was a place where 100s of people daily get their work done. it was all about thinking for the next place to optimize, adding features that help get things done faster — editor, playground, the small things that make shipping an agent feel less cursed.

spent time on SDKs — why should phone calls have all the fun, made the typescript SDKs & embeds to make voice stack available to developers. integration guides, realtime sockets, permissions, isolations for embeds, it was all a fun time. go to docs.sarvam.ai for trying it out.

then design system took over — we call it Tatva. late 2025 into early 2026 was mostly about stopping every feature from inventing its own button. pulled product UI onto shared primitives, wired tokens from figma, storybook, mcp tooling, docs that don’t lie about the api. boring on paper, but once it’s solid everyone (and later agents) just builds faster.

then the high tides.

feb 2026 — india ai summit in delhi. 14 releases in 14 days. not just voice agents either — product surfaces, platform bits, whatever had to be live for that week. compressed timeline, lots of parallel threads, somehow it landed at bharat mandapam.

july 2026 — epoch, our flagship conference. we opened what we’d been building to everyone — voice agents and the rest of the stack, live on indus.sarvam.ai. till then a lot of it was invite only; now anyone can show up and use it. we rebuilt how people create voice agents, bringing creation time from a few days down to under 10 minutes — genie in the loop, simpler authoring, deploy paths that don’t need a tour guide. the then/now below is that jump.

what i didn’t expect was how much room there’d be to work past the UI. at sarvam i got to sit with authentication, deployments, feature flags, partial rollouts — the stuff that decides whether a release actually reaches the right people.

still here

it’s 16 months since, still proudly at sarvam. none of what’s above was a solo run — designers, product, fellow engineers, the people who sat in the messy middle with the ideas and the tradeoffs. i just got to be in that room and ship with them.

the thing we’ve been shaping for a while is finally out with people. if you haven’t tried it yet, visit indus.sarvam.ai — and if something feels like it could be better, tell us. we’re always welcoming feedback.

that’s the year, compressed.

~ written by me and cursor