Hacking with AgentQL at Multimodal AI Agents Hackathon in San Francisco
Four teams reimagined how we interact with the web at the Multimodal AI Agents Hackathon. From voice-powered browsing to AI-generated cover letters, see how they used AgentQL to bring their ideas to life.
AgentQL was born from the belief that anyone should be able to ask the web for what they need using the power of language and AI. So when we partnered with Creators Corner to sponsor the Multimodal AI Agents Hackathon in San Francisco, we hoped to put AgentQL into the hands of builders from around the world.
Over the course of a two intense days, four teams shipped prototypes powered by AgentQL. From hands-free browsing to internship search automation, these projects didn't just use our API—they challenged how we use the web and imagined new possibilities. We were impressed by their creativity, curiosity, and scrappy execution.
And today, we're celebrating all four.
Grand Prize Winner: people-coming-over
What if your AI agent could see your room and redesign it for you?
Helen Koch and Kirill Igumenshchev built just that—a shopping agent that uses image recognition to suggest furniture and decor upgrades based on a photo of your space. And with AgentQL, it even handles checkout.
They built a full-stack LLM-native shopping flow using AgentQL, LlamaIndex, Samba Vision, ApertureDB, and more. The interface is clean and fast, the UX delightful. Honestly, we're still recovering from how cool this one is.
This project took home four awards! Check out the demo and details at Devpost or dive into the repo on Github.
Internship Hunt: Powered by Agno, Gemini, and AgentQL
Freshmen Stiven Triana and Hammon Dutra from Minerva University built an internship search assistant to save time and reduce friction in the job hunt. Upload your resume, get matched with internships, and receive custom-generated cover letters based on the requirements of each opportunity.
"We used AgentQL's REST API to check whether each internship required a cover letter, research statement, or why-us essay. Then we generated those automatically."
Their work demonstrates how AgentQL can surface structure from messy websites, enabling automation of tedious workflows. These students weren't just hacking—they were building tools they now use for their real internship applications. You can use it today or check out its source code on GitHub.
The judges at the event were so impressed, their project took second place! Check out their full writeup and demo on LinkedIn!
LookMaNoHands: Voice Browsing with AgentQL
Built by student Carl Kho exploring brain-computer interfaces, LookMaNoHands uses speech-to-text, AgentQL, and browser automation to navigate the web entirely by voice. No clicking. No scrolling. Just speaking.
"AgentQL was absolutely crucial... I used queryElements() extensively to locate UI elements using natural language. It's like having a superpower to control the browser with plain English!"
This project isn't just about accessibility—it's a step toward a future where our thoughts drive digital actions. Check out the writeup and code.
"What really surprised me about AgentQL was its robustness and adaptability to diverse website structures. The natural language queries genuinely seemed to understand the semantic context of web elements. Even when website layouts shifted slightly, AgentQL's intelligent selectors could still pinpoint the intended elements, making the automation remarkably resilient to UI changes. This showed me how far AI has come in understanding and interpreting the complexities of the live web – it's far more adaptable than I initially imagined."
Fixit: Real-Time Appliance Troubleshooting
Built by graduate students Prachi Sethi and Osheen Gupta, Fixit is an AI-powered assistant that helps users diagnose and resolve appliance issues using text, image, audio, or video inputs.
"We used AgentQL to extract the user guide directly from the brand's official site, once the appliance was identified. It gave us accurate, brand-specific instructions without needing to manually hunt down the right PDF."
Their biggest challenge was integrating multiple input modes—image, voice, text—and ensuring AgentQL could pull relevant data seamlessly. What surprised them?
"AgentQL let us write natural language queries instead of complex code. That made scraping guides much faster and easier than we expected."
If given more time, they'd expand the system to offer real-time, voice-guided troubleshooting and broader support across appliance brands.
Explore the Fixit GitHub repo.
"Special thanks to the AgentQL team for supporting us with guidance and updates during the hackathon. That kind of responsiveness really helped us move fast."
What we learned
These teams showed how AgentQL can be the glue connecting the live web and multimodal agents. They pushed our API to its limits—and helped us spot places we can improve.
Huge thanks to all the participants for teaching us so much and including AgentQL in their projects. Each team received access to our Professional Plan, and a few are continuing to build their ideas into real tools for students and users everywhere. If you're curious about how to get started with AgentQL, head over here.
—The TinyFish team building AgentQL