- unwind ai
- Posts
- Global AI Agent Hackathon Recap
Global AI Agent Hackathon Recap
6 Opensource AI Agent apps so good, We had to add them to Awesome LLM Apps GitHub Repo 🌟
Shubham Saboo & Gargi Gupta
June 28, 2025
Last month, we at Unwind AI wrapped up the Global Open Source AI Agent Hackathon, a focused build sprint for serious developers. No gimmicks, no wrappers. Just useful agent workflows, RAG systems, and tool-use workflows built with frameworks like Agno, Firecrawl, and Browser Use. With $25K+ in prizes and dozens of submissions, we saw a lot of interesting takes.
Now that the dust has settled, we’re featuring 6 standout projects. Each of these made it to our Awesome LLM Apps repo because they didn't just "demo the tech," they actually built something useful.
What The Hackathon Was About
This wasn’t just a prize hunt. The theme was simple: build agents that work. Think multi-agent coordination, tool-use, semantic exploration, RAG, local automation. Builders were free to use their stack, but we partnered with teams behind Agno, Mem0, Graphlit, Browser Use, and more to help devs plug into advanced capabilities faster.
Prizes were split across multiple tiers: 10 x $300, 10 x $500, 5 x $1K, 1 x $2K, and a $5K grand prize. But our personal shortlist wasn’t just about cash, it was about usefulness, originality, and the "damn, I could use that" factor.
So here’s how we picked our favorites. We looked for working demos, solid logic, useful outputs, and clean implementation. If your app could plug into a real-world use case today (or in a few weeks), it got our attention. If it was also well-packaged and made smart use of agents, that sealed the deal.
Our Favourite Projects
Courtesy: Guanming Wang
Likeminds is a visual semantic network tool that helps you discover and connect with like-minded people across different platforms like LinkedIn, Twitter, GitHub, and more. Instead of relying on follower/following graphs, it builds a graph using semantic similarity based on areas like background, affiliations, and expertise.
You start by submitting your social media links and other documents, which are processed by a personalized agent. The agent performs personality analysis and builds a knowledge graph of your experience. Based on your inputs, it can then scan public data to find individuals who share common interests, institutions, research, or career paths.
The results are shown as an interactive graph, where you can click on nodes to go deeper. Every profile includes reasoning for why the match was made, and your agent can even draft personalized messages to start the conversation. Need more info on someone? A deep research agent digs into their achievements, background, and online presence.
Tech: Agno, GPT-4o, Gemini, Perplexity, Google APIs, LinkedIn API, X API, Instagram API, React
Courtesy: Amina Javaid
AI Speech Trainer is your AI-based public speaking coach. It combines video, audio, and text inputs to help you improve how you speak, present, and carry yourself during presentations. Whether it's a job interview, pitch, or conference talk—this app offers detailed coaching feedback.
You upload a recording of your speech. Then, agents get to work:
A facial agent tracks your expressions and eye contact.
A vocal agent transcribes and evaluates tone, pitch, and pace.
A content agent checks your structure, grammar, and clarity.
A coordinator agent pulls everything together and scores your performance using a simple rubric.
The feedback is clear, targeted, and actionable—showing you where you did well and where you need work. Built with a multi-agent setup via Agno, the project is modular and ready for future additions like live coaching, dashboards, and replay with an AI avatar.
Tech: Agno (multi-agent), OpenCV, Mediapipe, DeepFace, Together Llama-3.3 70B, Streamlit, FastAPI
Courtesy: Jeomon George
Windows-Use is a local AI agent that interacts directly with your Windows desktop. It can click buttons, open apps, type into fields, and execute shell commands - basically automating any action a user might do manually. But unlike most GUI automation, this agent uses system APIs instead of computer vision, making it faster and more stable.
You can run it in GUI or CLI mode. It supports real-time task execution and even reads from documents (PDFs, URLs) to figure out what to do. Want to automate setting up a VM or installing software based on a guide? Feed it the documentation and it follows along.
A lot of thought went into the agent structure, including memory, UI state tracking, and error handling. It's still under development but already supports real system-level interaction.
Tech: Agno, Gemini 2.0 Flash, Claude Sonnet 4.0, PyQt6, Groq (Whisper + TTS)
⚠️ Pro tip: run it in a sandbox. It has real power.
Courtesy: Arun Arumugam
Beifong is a personal podcast studio powered by agents. You connect your trusted content sources - RSS feeds, Facebook, X - and Beifong pulls in new content, analyzes it, and turns it into fully produced podcast episodes with voiceovers and banners.
It comes with an orchestration layer built on Agno. Individual processors handle content classification, sentiment analysis, script writing, TTS voice synthesis, and visual asset creation. There's also a browser agent via BrowserUse to crawl the web for new articles or social content.
You can run it via a browser interface, call it through APIs, or set it on autopilot with a scheduler. It also supports Slack integration, where you can talk to your agent directly inside a Slack thread. You can even write your own tools and plug them in.
Tech: Agno, Playwright, Redis, OpenAI, ElevenLabs, FastAPI, FAISS, Browser Use
Courtesy: cmena-dev
TubeWarden is a Chrome extension powered by an AI agent that filters your YouTube experience based on your personal standards. It lets you rate and filter videos by safety, clarity, relevance, and more, helping you avoid misinformation or low-quality content.
When you browse YouTube, the extension detects video URLs and sends them to the backend agent. Based on your selected categories and thresholds, it scores each video and overlays the result on the page. You can automate responses like hiding certain videos or flagging them as "not interested."
The tool is fully user-configurable. You choose what you care about and define how strict the filters are. This gives users control over their algorithmic experience, not the other way around.
Tech: Agno, GPT-4o, Groq, FastAPI, Chrome Extension
Courtesy: Amit Wani
TripCraft AI is your full travel planning assistant. Instead of spending hours comparing flights, hotels, and itineraries, just tell it your destination, dates, and budget—and it does the rest.
It runs 6 agents in parallel:
A destination agent researches attractions, activities, and spots.
A hotel agent finds good stay options with filters.
A dining agent lists restaurants matching your preferences.
A flight agent searches air travel routes and timings.
A budget agent keeps cost under control.
An itinerary agent puts everything into a daily schedule.
The result is a complete plan with clear bookings, activity flow, and budget optimization. It's personal, fast, and removes the stress from travel prep.
Tech: Agno, Gemini, Firecrawl, Exa, Google Flights API, Kayak API, FastAPI, React/Next.js
Final Thoughts
Every one of these projects used agents the way they’re meant to be used: for delegation, automation, and specialized coordination. They’re not perfect, but they’re real. And they show how far we’ve come from basic chatbots.
Big shoutout to all 300+ participants and 30 final winners. If you missed this one, stay tuned. We’ll be back soon with another round.
In the meantime, check out the full Awesome LLM Apps repo to learn about 75+ AI Agents, MCP and RAG apps, go fork and build something useful.
We share hands-on tutorials 2-3 times a week, to help you stay ahead in the world of AI. If you're serious about leveling up your AI skills and staying ahead of the curve, subscribe now and be the first to access our latest tutorials.
Reply