A set booth. A set size. A set time.
Could we meet people our own way, beyond all that?
AI Everywhere. In the future Nota AI envisions, there is always a moment where more people and more technology come together. Over two months this spring, Nota AI set out to create that moment for itself through 2026 KCC and 2026 ICML.
From Jeju Island to Seoul, from booths and competition halls to our own office. Across these different spaces, we kept searching for an answer to that question. At KCC, our answer was to put our technology directly into the hands of people outside our own team. At ICML, it was to open our own doors and invite people in. This post shares what Nota AI tried for the first time.
From explaining to letting go (KCC 2026)
Nota AI co-hosted the "On-device AI Optimization Competition" with the Korean Institute of Information Scientists and Engineers (KIISE). It was the first time Nota AI ran a competition at KCC instead of a booth, and the first time we stood as an official co-host alongside the society itself. After weeks of applications, roughly 20 teams applied, and following two rounds, 7 teams made it to the final round in Jeju. The grand prize team boosted a Llama model's response speed by more than 8x on a Raspberry Pi, while cutting memory usage by the same margin.
What Nota AI focused on in this competition wasn't explaining, it was preparing. What does that mean? We handed NetsPresso to participants who had never formally worked with LLM optimization, gave them just 30 minutes of training to absorb that complex pipeline on their own, and then left the outcome entirely in their hands. Instead of showing off what we had built, we stepped back and simply provided the process, so that results we hadn't made ourselves could emerge. That, in itself, was the most meaningful part of this competition. It was a chance to test, in a different way, whether our product could hold up in unfamiliar hands. It was also a chance to gauge whether AI could reach beyond expert tools and into the hands of many more people.
From attending to inviting (ICML 2026)
ICML, widely regarded as one of the most prestigious conferences in AI/ML. Separate from our booth there, Nota AI created two more spaces this time: Korea Efficient Night, a networking event, and Open Office Days, an office open house. About 100 people attended the networking event, and even more visited during Open Office Days. It was the first time Nota AI, which had always only attended conferences, took on the role of host instead.
In preparing these two events, what mattered to Nota AI wasn't how many people showed up, but how deep the conversations went. For Efficient Night, we analyzed attendees' background information in advance and assigned each table a relevant keyword. Some tables were grouped by specific technical topics, like Quantization/Pruning, Inference Optimization, and Model Compiler/IR. Others were grouped by emerging research trends, like Multimodal Foundation Models, On-device VLMs, and Agentic Model Optimization. Between them sat one table dedicated to mentoring and career conversations.
None of these tables were led by someone from communications or recruiting. Every one was moderated by the actual researchers and engineers doing that work. Open Office Days followed the same principle. Rather than a simple space tour, we made sure visitors could see our published papers, our culture, and our actual products up close, and put questions directly to the people building them, right there on the spot. AI was the one thing that brought everyone to the same table, whoever they were and whatever brought them there, and that was the whole point. So whether someone was curious about our technology, our company, or our products, we made sure that whichever direction they came from, they could get their answer in that same room.
So what did we take away from this?
Relationships.
What we gained from KCC wasn't a one-off connection. Nota AI had shown up at KCC with a booth for years, and that history is what made us a co-host standing alongside the society itself this time. After the competition ended, several professors who had watched the event reached out, asking to bring NetsPresso hands-on exercises into their regular coursework. That is a far heavier signal than a single hire or a small bump in brand awareness. It means a pathway now exists for students, semester after semester, class after class, to encounter Nota AI's tools.
The relationships we built at ICML had a different texture. The tables at Efficient Night held everyone from students just starting out in research, to overseas researchers in Seoul for ICML, to representatives from global companies. With that range came an equally wide range of questions, and having practitioners seated at all twelve tables meant we could actually answer them. Showing visitors the people who tackle our everyday problems, and the very room we work in every day, ultimately led to real hiring conversations too.
From students to researchers at global companies, from a domestic academic conference to a world-class one. The relationships Nota AI built this time stretched in many directions at once.
Conviction.
The two events gave us different kinds of conviction, and placing them side by side made each one mean more.
The KCC competition gave us conviction that our tools work even in the hands of beginners. Students who had never formally touched LLM optimization absorbed the entire pipeline in just 30 minutes of training, and produced strong results in ways we hadn't designed ourselves. It was a sign that NetsPresso isn't just a tool for experts.
Over the same period, the ICML stage gave us conviction of a different kind. We placed third out of 40 teams worldwide in the Efficient Qwen Competition, and had two papers on MoE quantization accepted into the AdaptFM workshop, whose organizing committee includes researchers from Amazon and Meta. What mattered more than the ranking or the acceptance itself was that these results came from a stage watched by researchers around the world. It was a moment that confirmed Nota AI's research holds up not by our own internal standard, but by a global one.
Teamwork.
What we took away from this wasn't just the two things above. It was also months spent with teams from marketing, research, product, and recruiting all moving together around the same events, building a real sense of what it means to work as one team. There were moments when Nota AI's own people huddled intensely backstage to pick a winner among genuinely strong entries, and moments when an unexpected surge of visitors meant scrambling to bring in extra hands on the spot. Every one of these formats, the booth, the competition, the invitation events, was something Nota AI was doing for the first time, so there was no playbook. We learned a lot by running into problems as they came.
But none of that friction was a one-time cost. The routing and staffing we scrambled to figure out on the day became a checklist for the next event, and a shared sense of which team should own what naturally took shape among the people involved. The instinct for collaboration we built by pulling off two new kinds of events in two months will, we believe, quietly support Nota AI every time we take on something bigger going forward.
.
.
.
In closing
On the wall of our office lounge hangs the phrase, "Stay Curious, Stay Connected." It carries the hope that Nota AI's technology, which started from curiosity, will become a Good Influence, and eventually reach our vision of AI Everywhere.
Nota AI didn't wait behind a booth for people to come to us. We reached out, opened our doors, and met curious people halfway. To the question we started with, whether we could meet people our own way, beyond a set booth, a set size, a set time, this year's KCC and ICML gave us one answer.
Just like the words on our lounge wall, we expect Nota AI will keep reaching more people from here.
