Inspiration
As a physics teacher in Brazil, I work with students who design, simulate, build, and launch educational rockets. Over time, each project generates valuable information: mass, dimensions, center of gravity (CG), center of pressure (CP), stability, motor configuration, simulation results, and real launch performance.
The problem is that much of this knowledge is lost between projects. Rocket Brain AI was created to give educational rocketry a persistent AI memory.
What it does
Rocket Brain AI is an AI agent designed to help students and educators analyze and learn from rocket projects.
The system is designed to:
- analyze rocket design parameters such as mass, diameter, CG and CP;
- reason about stability and simulated performance;
- store information from previous rocket projects;
- retrieve previous experiments and launch results;
- compare simulation data with real-world results;
- help students understand why one design may perform differently from another.
Instead of treating every question as a new conversation, Rocket Brain AI is being built to remember previous projects and turn accumulated experiments into reusable knowledge.
How we built it
The prototype was developed in Python.
We used AWS Strands Agents to build the AI agent and Amazon Bedrock as the foundation-model infrastructure.
Cognee is used to explore persistent memory and knowledge representation for rocket projects. Project data can be processed into a memory layer that can later support retrieval and comparison across experiments.
During the hackathon, we successfully built the agent architecture, connected the development environment to Amazon Bedrock, and processed an example rocket project through Cognee.
Challenges we ran into
One major challenge was configuring a new AWS account and Amazon Bedrock access during the hackathon. Account verification temporarily limited model inference.
We also explored how Cognee handles local knowledge extraction and learned how the memory pipeline requires both knowledge extraction and an LLM configuration for some search operations.
These challenges helped us better understand how agent reasoning, model infrastructure, and persistent memory can work together.
Accomplishments that we're proud of
We transformed a real educational problem into an AI-agent architecture during the hackathon.
Rocket Brain AI combines physics education, experimental rocketry, AI agents, and persistent memory.
Most importantly, the project is based on a real educational workflow: students design rockets, simulate them, build them, launch them, collect results, and improve the next generation of designs.
What we learned
We learned how to build an AI agent with Strands Agents, connect Python applications to Amazon Bedrock, and experiment with Cognee as a memory and knowledge layer.
We also learned that persistent memory can make educational AI more useful because experiments from previous students can become knowledge for future projects.
What's next for Rocket Brain AI
Next, we plan to:
- complete the integration between the Strands agent and Cognee memory;
- import OpenRocket simulation data automatically;
- calculate and validate rocket stability metrics;
- compare predicted and real launch performance;
- build a visual interface for students and teachers;
- create a growing knowledge base from real student rocket launches.
The long-term vision is to create an intelligent educational rocket laboratory where every launch teaches the AI something that can help improve the next one.