A look at the top teams from OpenInfer’s first builder hackathon — and what they built on distributed inference.
This June, OpenInfer hosted its first builder-focused AI hackathon, bringing together teams in San Francisco and remotely to build creative, agent-driven applications on distributed inference.
In a single event, participants turned ambitious ideas into working applications: from multi-agent brand-compliance engines and natural-language design tools to sports tracking, ad intelligence, and agent-generated documents. Many of the projects put OpenInfer’s inference capabilities to work, running multiple AI tasks and agents in parallel while balancing performance, latency, and cost.
Judges evaluated each project on design, user experience, and execution. The field was so competitive that the final rankings came down to fractions of a point.
Here are the builders who took the top spots and what they built.
1st Place: DoItAll “On the Fly”
Sujal Thapa and team
What if editing a website, product, or design were as simple as circling the part you want changed and saying what you want? That’s the idea behind On the Fly, the project that took first place.
On the Fly lets non-technical users draw a freeform region on any website, product, or design and give plain-language instructions — the agent focuses its edits on exactly that section, nothing else. Behind the scenes, persistent memory cross-references your prior designs and styles, so the tool gets sharper about your taste with every edit. It’s a simple interaction model hiding a genuinely hard technical problem, and the judges gave it top marks across the board — design, experience, and execution.
The team brings real agent-building pedigree: Sujal built AI agents at Kiwi Charge, and his teammates come from Bell and a stealth startup working on agents and AI inference.
See it in action: On the Fly demo
2nd Place: PureConstruct AI
Miguel Velazquez and Nyi Htet
Construction is one of the world’s largest industries — and one of the least digitized. PureConstruct AI, built in person in San Francisco by Miguel Velazquez and Nyi Htet, points agentic AI squarely at that gap.
The duo impressed the judges with some of the strongest execution of the entire event, earning the second spot overall in a tightly contested field. It’s the kind of project we love to see at a hackathon: a real industry, a real problem, and a working demo built over a weekend.
See it in action: PureConstruct AI demo
Spotlight: BrandOps
Krish Patel and Ved Dabhi
Every creative team knows the pain of brand review: does this asset match the guidelines, is it accessible, does it create legal risk, does it fit the audience? BrandOps answers all of those questions at once.
Upload your brand guidelines and creative assets, and a team of specialized agents evaluates brand consistency, accessibility, legal risk, visual hierarchy, and audience fit — while tracking cost, latency, and each agent’s reasoning along the way. Built by SJSU computer-science students Krish Patel and Ved Dabhi, BrandOps was designed explicitly to showcase what distributed inference can do when multiple agents work a problem in parallel.
See it in action: BrandOps demo
What’s Next
As part of their prize, the winning teams are presenting their projects to senior representatives from Cota Capital, a San Francisco-based technology investment firm that partners with founders building enterprise and deep-tech companies.
Thank you to every team that participated, to our judges, and to everyone who helped make the first OpenInfer Hackathon happen. The creativity and quality across every submission made this an easy event to be proud of — and a hard one to judge. We can’t wait to see where these builders take their projects next.
Follow OpenInfer for more from our builder community.