Inspiration
I was frustrated with how science is usually taught mostly theoretical, with limited access to real labs. Many students never get the chance to actually experiment, especially in environments where lab equipment, teachers, or even stable internet access are unavailable. I wanted to build something that removes these barriers entirely.
The idea behind ExperimentAI is simple: even with just a basic device, anyone should be able to explore physics, build circuits, simulate systems, and truly understand concepts with the help of AI. Instead of building another edtech platform, I set out to create a real virtual laboratory one that runs on your device, works offline, and feels like having an intelligent scientific guide beside you at all times.
What it does
ExperimentAI is a local-first, AI-powered virtual lab that allows users to perform interactive physics, electrical, and astronomy experiments directly in the browser. Users can simulate systems, collect real-time data, and receive intelligent, context-aware explanations during and after experiments.
The platform works even without internet connectivity, storing all experiment data locally and syncing seamlessly when a connection is available. It also generates structured lab reports with AI-driven insights, turning every experiment into a complete learning experience. The goal is to make science not just accessible, but deeply interactive and intuitive.
How we built it
I built ExperimentAI using a modern full-stack architecture focused on performance, interactivity, and resilience. The frontend is designed for smooth simulations and a clean, professional SaaS-style experience. Supabase is used for backend services and authentication.
PowerSync powers the local-first architecture, enabling instant local data storage with seamless background synchronization to the backend. For AI capabilities, I used Mastra to orchestrate intelligent workflows such as experiment analysis and reasoning, and Tavily to enhance responses with relevant scientific knowledge. The system integrates simulations, AI decision-making, and real-time sync into a unified platform.
Challenges we ran into
One of the biggest challenges was building a reliable local-first system while maintaining consistent real-time synchronization. Handling edge cases between offline and online states required careful design to avoid data conflicts and ensure a smooth user experience.
Another challenge was making AI responses genuinely useful in a scientific context. It required structuring experiment data and workflows so the AI could reason based on actual results instead of generating generic explanations. Ensuring simulation accuracy while maintaining performance in the browser was also complex.
Additionally, integrating multiple systems authentication, syncing, AI workflows, and simulations introduced challenges around data flow and state management. Debugging these interactions was time-consuming, but it ultimately led to a more robust and scalable system.
Log in or sign up for Devpost to join the conversation.