⏺ Inspiration
Growing up, I was always drawn to word search puzzles in the back of science magazines. There was something magical about hunting for hidden words while learning about fascinating topics like DNA structure or photosynthesis. But I always wished the puzzles were more dynamic — what if I could create a puzzle about any biology topic I was curious about? That childhood dream combined with my passion for AI led me to build BioSearcherAI. I wanted to create something that captures that same excitement of discovery, but powered by cutting-edge AI agents that can research any topic and create personalized learning experiences.
🧩 What it does
BioSearcherAI transforms any biology topic into an engaging word search puzzle using sophisticated AI. Users simply enter a topic like “parts of the kidney” and our LangGraph agents spring into action — researching Wikipedia, extracting relevant biological terms, and strategically adding distractor words from related fields. Claude Sonnet 4.5 then generates a custom 15x15 grid where players hunt for hidden words without knowing what they're looking for. After submission, the game reveals all target words, shows what was missed, and provides educational context for deeper learning.
⚙️ How we built it
I architected this as a modern full-stack application with a clear separation of concerns. The frontend is a beautiful Next.js app with React and Tailwind CSS, featuring an interactive word search grid with drag-to-select functionality.
The real magic happens in the Python backend, where I implemented sophisticated LangGraph agents that create multi-step workflows:
- One agent researches Wikipedia content
- Another finds strategic distractor terms
- A final agent coordinates puzzle generation using Claude’s advanced
language capabilities
The two systems communicate through a clean Flask API, creating a seamless user experience powered by state-of-the-art AI.
🧠 Challenges we ran into
The biggest challenge was getting the LangGraph agents to work reliably with the Claude API. Initially, I struggled with authentication errors and model configuration issues that took hours to debug.
Wikipedia’s disambiguation pages also caused unexpected errors when researching certain topics. Additionally, creating a responsive word search grid that works smoothly on both desktop and mobile required careful attention to touch events and drag interactions.
Balancing the complexity of the AI backend with a simple, intuitive user interface was an ongoing challenge throughout development.
🏆 Accomplishments that we're proud of
I'm most proud of creating a genuinely educational tool that combines my love for puzzles with cutting-edge AI technology. The LangGraph workflow is sophisticated enough to research any biology topic intelligently, yet the user experience remains delightfully simple.
The system successfully generates challenging puzzles by mixing relevant terms with clever distractors, creating an experience that's both fun and educational.
Seeing the detailed console logs show the AI agents working together — researching, analyzing, and creating — feels like watching a glimpse of the future of personalized education.
💡 What we learned
This project taught me the power of agent-based AI architectures.
Working with LangGraph showed me how breaking complex tasks into
specialized agents creates more reliable and maintainable systems than
monolithic approaches.
Most importantly, I learned that the best educational tools combine sophisticated technology with simple, engaging user experiences.
🚀 What's next for BioSearcherAI
The future possibilities are endless! I'm excited about expanding beyond biology to create subject-specific puzzle generators for chemistry, physics, and history.
I want to add:
- Difficulty levels
- Collaborative multiplayer modes
- Adaptive learning that adjusts based on user performance
The ultimate goal is to create a comprehensive learning ecosystem where curiosity drives personalized, AI-powered educational experiences that make learning as addictive as my childhood puzzle obsession.
Log in or sign up for Devpost to join the conversation.