July 13, 2026 · 8 min read
300 Builders, 86 AI Projects, One Weekend: Inside the First Codex Community Hackathon Pune
Most hackathons run on caffeine and hope. This one ran on caffeine, hope, and a platform that actually got out of the builders' way. On June 14, 2026, Pune hosted its first OpenAI Codex Community Hackathon, a single, intense day of shifting from "writing code" to "orchestrating agents." 1500 builders registered, out of which 300 builders were selected and. 86 projects were shipped. And the whole thing registration, AI-assisted ideation, submissions, and judging ran on Loops House.
Here's the story of what got built, who won, what builders told us afterward, and what we're fixing because of it. But first, what exactly is Loops House? “Loops House is an AI-native hackathon platform where builders ship, prove, and compete using real AI tools sponsor docs installed directly into a builder's IDE, AI-assisted ideation, and an AI Judge that evaluates every submission on evidence, not just pitch quality but the pitches in detail. In June 2026, Pune hosted its first OpenAI Codex Community Hackathon by OpenAI Codex Ambassadors together with LocalDev run end-to-end on Loops House. Here's what happened, what builders told us afterward, and what we're changing because of it.”
The event, by the numbers
Pune's first OpenAI Codex Community Hackathon asked builders to move past the chat interface and architect autonomous agents not simple chatbots, but tools that build, debug, and orchestrate software on their own. The event ran across five build directions: Agentic Coding, UX for Agentic Applications, Multimodal Intelligence, Domain Agents, and Building Evals.
300 builders registered
86 projects submitted
$9,500 in prize pool, sponsored by Codex
We received post-event feedback from over 90% of participating teams, and the majority of response was positive. Most participants highlighted the overall quality of the experience, with many praising the mentorship, community, and program structure. Around half also shared thoughtful suggestions on how we can make future editions even better, while only a handful encountered issues that meaningfully impacted their experience. The feedback has given us a strong foundation to build on, and we're excited to incorporate these learnings into the next iteration of the program.
The setup
The event was organised by OpenAI Codex Ambassadors together with LocalDev, Pune's community of local builders, and hosted at the Infosys Training Centre in Pune. Codex - OpenAI's AI coding agent, was the sponsor stack, and the brief was simple to say and hard to do: move past the chat interface and build agents that can actually build, debug, and orchestrate software on their own.
Five build directions were on the table:
Agentic Coding dev tools that get real leverage out of Codex as a coding agent
UX for Agentic Applications interfaces designed for a world where an agent is doing the work
Multimodal Intelligence applications reasoning across more than just text
Domain Agents vertical agents built for real-world constraints
Building Evals tooling to measure and improve how well an agent actually performs
Teams of up to 4 (solo builders welcome), one day, and a $9,500 prize pool on the line.
The winners
Judging ran through Loops House's AI Judge, reading actual repositories, not slide decks with human reviewers confirming the final calls. Here's who took home the bounty:
🥇 1st Place: $5,000 Air Secure An agentic drone security analyst that turns fixed-camera footage into searchable events, deterministic alerts, and real-time Telegram evidence. Security tooling that actually tells you what happened, not just that something did.
🥈 2nd Place: $3,000 FORGE Forge is an autonomous engineering org in a box. You describe what you want in plain English; a swarm of Codex agents builds it in parallel, an eval gate scores and kills the weak attempts, the best parts merge into one product that ships itself to a live URL and when it breaks, Forge heals it without a human in the room.
🥉 3rd Place: $1,500 RedShield_Agent An automated red-teaming platform that stress-tests LLM applications and hands back verified prompt-hardening recommendations builders using agents to make other agents safer.
Rounding out the Top 10: SentinelOps (approval-gated DevOps automation for Codex), Touchstone (a data lie detector that proves every number and writes its own checks), ReproPatch Studio (turns bug reports into verified, PR-ready patches), AgentCI, CollabJam Studio (a Git-native music studio where AI agents compose in parallel), SkySwarm, and WhiteBoard (turning a rough idea into an implementation-ready architecture plan).
More Projects That Stood Out
86 submissions in one day means a lot of range, and a lot of personality. A few that deserve a shoutout even though the prize money went elsewhere:
SkillBridge-AI built an AI career coach that reads your resume, scans your GitHub, checks live market signals, and builds an actual roadmap to get you hired "not a generic one," as the team put it.
ArchiteX wins the award we didn't officially give out for funniest one-liner on the platform: "Turning your illegible, caffeine-fueled whiteboard scribbles into a functional codebase before the judges realise we haven't slept." Relatable content for anyone who has ever hackathon'd.
PunaRaksha "a city that listens, warns, and protects" and VibeGit, a self-correcting sandbox for safer "vibe coding," both showed up in the general project feed and stuck with us.
Browse everything the room built →
Builders loved the AI tooling on Loops
The AI-native workflow emerged as one of the strongest themes in post-event feedback. Builders repeatedly highlighted how the hackathon-specific Codex skill preloaded with the event's tracks, judging criteria, and planning scaffolds helped them move from an idea to a structured project much faster. Features like /ideate and /refine made it easier to sharpen pitches, iterate on concepts, and align with sponsor requirements, while persistent conversation artifacts meant teams could pick up where they left off without losing context.
"Loops House feels like a unified platform for hosting, evaluating, and ideating on hackathon projects, powered by AI-driven tools. It did help me refine my ideas under hackathon constraints."
— Rutav Desai, Codex Community Hackathon participant
Beyond the AI features, builders consistently praised the submission experience. The three-step submission flow made it easy to save progress and return later, while bringing event information, sponsor resources, timelines, and submissions into a single workspace reduced context switching and let teams stay focused on building.
Participants also appreciated the clean interface, fast registration and check-in process, reliable platform performance throughout the event, and the live project feed, which gave teams visibility into what others were building in real time.
These themes were consistent across the post-event survey. Of the 85 teams that completed the survey, 79 shared open-ended feedback. Most described a positive experience, many paired their feedback with thoughtful suggestions for improvement, and only a small number reported issues that significantly affected their participation.
Always be iterating
Even inside overwhelmingly positive feedback, the same handful of requests kept surfacing and they clustered into three clear areas, the honest list of what didn't work, and what we're doing about it.
A submission checklist is coming. By far the most-requested feature was simple, a clear view of what's done and what's still missing before the deadline, plus a live countdown instead of guessing. Loud and clear, and it's now a roadmap priority.
Judging is getting more transparent. Several teams weren't sure exactly how they'd be scored. We're building a judge-view preview so builders can see their submission the way the AI Judge sees it, before they lock it in, plus sharing examples of strong past submissions earlier next time.
Collaboration and notifications need work. Right now a submission can feel like it has one owner. Real multi-member co-editing, plus live deadline and announcement alerts (Discord/Slack integration included), are both in motion.
A smaller number of teams also ran into specific friction: a permissions error that temporarily blocked project updates for a few teams, one lost in-progress co-pilot session during navigation, difficulty deleting a submission or adding teammates after a submission had already gone in, and some rough edges representing CLI-first, local-only tools on a platform built around live app URLs. We're treating each of these as a concrete bug or gap to close, not just feedback to note.
Inviting all builders
The core of what Loops House is built to do get builders from idea to a judged, evidence-backed submission with as little friction as possible clearly landed. The requests that came back weren't about the fundamentals; they were about tightening visibility into status, judging, and collaboration around that core flow. That's a good problem to have, and it's exactly the kind of feedback loop the platform is designed to close quickly.
Frequently asked questions
Is it free to participate in a hackathon on Loops House? Yes, hackathons on Loops House are always free for builders to join. Prize pools, sponsor tracks, and platform costs are covered by event hosts and sponsors, not participants.
How do I find and join other hackathons on Loops House? Head to Explore Events to see live and upcoming hackathons. Each event page lists its build tracks, prize pool, and submission requirements, and registration happens right on the platform.
Can I participate solo, or do I need a team? Both are usually welcome. Team size limits vary by event (this one allowed up to 4), but solo builders are almost always an option check the specific event page for its rules.
What could builders build at the event? It depends on the theme of the hackathon, for the codex hackathon we had five build directions were open: Agentic Coding, UX for Agentic Applications, Multimodal Intelligence, Domain Agents, and Building Evals all centered on Codex as the underlying agentic coding tool.
How was the event run on Loops House? Loops House handled registration, a Codex-specific AI skill installed into each builder's IDE for ideation and scaffolding, real-time project tracking, and AI Judge evaluation of every submission based on actual repository evidence.
How were projects judged? Every submission was evaluated by Loops House's AI Judge, which reads the actual project repository and evaluates it against the sponsor's criteria, with human reviewers confirming final placements.
Do I need to install anything to participate? Just one terminal command. Every event installs a hackathon-specific Skill directly into your IDE Cursor, Claude Code, Codex, VSCode, or any agent on the open .agents standard preloaded with that event's tracks, sponsor docs, and judging criteria.
How are projects judged? Every submission goes through Loops House's AI Judge, which reads the actual repository and evaluates it against the event's bounty criteria not just the pitch. Human reviewers confirm final placements.
I am a hackathon organizer, how can I use Loops house for an upcoming hackathon? Yes. See how a Loops hackathon works or reach out at contact@loops.house to book a demo. Loops supports everything from a first-time community event to a multi-sponsor developer conference.
Can my company or community partner with Loops House as a sponsor? Yes, Loops works with sponsors, ecosystems, and enterprise teams to run hackathons, put real projects in front of their team, and surface top builders. Reach out at contact@loops.house to talk through what that could look like.
Want to run an event like this one? See how a Loops hackathon works, or explore live hackathons on Loops including the original Codex Community Hackathon Pune event page and the full gallery of projects builders shipped.
