310 People, 7 Hours, One Big Hands-On Day with NVIDIA Nemotron 3 Super in Bengaluru

There's a particular kind of energy you only get when 300+ people show up on a weekend, laptops open, willing to spend seven hours on a single model family. That's what yesterday looked like at the Amadeus IT Group office in Bengaluru, where we hosted the NVIDIA Nemotron 3 Super Meetup and I'm still processing how well it came together.

This is my recap as one of the organizers. Not a marketing summary. The honest version ~ what worked, who showed up, the small things that mattered.

How the day actually unfolded

We kicked things off around 9:00–9:30 AM. People were already trickling in by 8:45, which is rare for a Bengaluru meetup ~ usually the first hour is half empty. Not this time.

Megh Makwana from NVIDIA opened with the workshop session. And when I say workshop, I mean a real one — laptops out, terminals open, people actually running things. He walked through the Nemotron Super architecture and got everyone hands-on with the model. The screen behind him had tmux panes split four ways, with 01_basic_inference.ipynb, 02_thinking.ipynb, 03_performance.ipynb and the launch scripts all visible at once. You could feel the room lock in.

Right after Megh, Utkarsh Uppal from NVIDIA picked up the thread and went deeper. By the time his session ended, hands were going up faster than we could route the mic.

Then we broke for lunch. More on that in a bit, because the lunch story is genuinely one of my favourite parts of the day.

After lunch, the Sarvam AI team came in for a 30-minute talk. Their architecture deep-dive — Sarvam 30B vs 105B, GQA vs MLA attention, 128 sparse experts, MoE routing stability — was the kind of content you usually only get at a paid conference. The room was packed and quiet. People were taking notes.

Next up was Gaurav Tyagi from the Amadeus team. Gaurav didn't do slides — he did a live demo. He took an existing flight booking flow and showed how an agentic AI layer could modify and orchestrate it on the fly. Live demos at meetups are always a coin flip. This one landed.

Then Manjunath Janardhan closed the technical sessions with a talk on scaling to 120 billion parameters — the kind of session that makes you re-think what "large" even means in 2026.

I jumped on stage near the end with Megh to announce something I'm genuinely excited about: a four-week contest for the community. Build a project, write a blog post, ship something with Nemotron Super — there are prizes, and more importantly, there's a real chance to get your work seen by NVIDIA. I also walked through the Awesome Nemotron repository I've been curating — agentic AI examples, deployment recipes, integrations — and pointed people to where they can contribute.

We wrapped at 4:15 PM. And here's the thing about Bengaluru meetups that always gets me — people stayed. They didn't bolt for the parking lot. They hung around, asked questions, exchanged GitHub handles, asked when the slides would be up (they're coming, I promise).

The lunch story

I have to give a shoutout to our food vendor that catered the food. They delivered 350+ lunch boxes at 12:30 PM sharp, on time, no chaos. If you've ever organized food for a 300+ event in Bengaluru, you know how rare this is.

The food was actually delicious. Multiple people came up to me afterwards specifically to say so. Small thing, big impact.

The venue and the people who made it happen

Amadeus IT Group's Bengaluru office turned out to be the perfect venue. Seven or eight screens spread across the hall meant nobody ~ even folks at the back tables had to squint to follow along during the workshop. The AV team kept things running smooth, and that matters more than people realize.

A few people I have to call out by name, because nothing about this day worked without them:

• Savitha Pareek (NVIDIA) — co-host of the entire day, kept the energy up, kept the schedule on track. She also wrote a beautiful LinkedIn recap that captured the spirit of it.

• Kavita Aroor (NVIDIA) — coordination behind the scenes that nobody saw but everybody benefited from.

• Satyam Soni — co-hosted alongside Savitha and ran point on a hundred small things during the day.

• Koti Vellanki — community member who jumped in wherever needed, no questions asked.

• Raveendiran RR — same. Showed up, did the work, didn't ask for credit.

• The full Amadeus Labs team — for opening the doors, the venue, and the support.

What I'm taking away

A few things stuck with me afterwards:

The hands-on format works. People don't want another panel discussion. They want a terminal, a model endpoint, and a problem to solve. We had all three, and the room responded.

The Bengaluru AI community has changed. Five years ago, a meetup like this would have pulled 60 people. We had over 300, and we could have had more if the venue allowed it. The hunger to build with frontier models is real, and it's local.

The community is the multiplier. NVIDIA brought the tech. Amadeus brought the venue. Sarvam and Amadeus brought the talks. But the energy in the room — the questions, the side conversations, the laptops staying open through lunch — that came from the Collabnix community and the broader Bengaluru builder scene. That part doesn't scale with a marketing budget. It scales with trust built over years.

What's next

If you attended and you want to:

1. Join the contest — four weeks, ship a project or a blog post with Nemotron Super, win recognition (and prizes).
2. Star and contribute to Awesome Nemotron — that's where I'm collecting community projects, agentic examples, and deployment guides.
3. Slides — they're coming. Multiple people asked. We'll publish them through Collabnix and the NVIDIA Developer channels.

If you missed it — there will be more. The Collabnix community runs these regularly across Docker, Kubernetes, and now increasingly Agentic AI. Join us at collabnix.com or the Slack/Discord (~17,000+ of us already there).

To everyone who showed up, asked questions, opened their laptops, stayed past 4 PM, and made this what it was — thank you. This is what community looks like when it's working.

See you at the next one.