PhysicsLens

Inspiration

I'm a junior at highschool and my physics teacher is not really good. Most of our class don't understand the concepts and find them difficult. I realized the problem wasn’t physics itself, it was how it was being taught. We needed to see forces, not just calculate them.

What it does

The user uploads a video or records one and we select 3 frames, draw vector forces on them and then provide an explanation according to the user level.

How we built it

PhysicsLens combines computer vision + LLM: Architecture: Video Input --> YOLO Object Detection --> Frame Selection --> Gemini-3 Physics Reasoning --> JSON Parsing --> OpenCV Vector Overlay

YOLO detects objects involved in the physical interaction
Gemini-3 reasons about forces, directions, and magnitudes and tells OpenCV exactly where to draw vectors
OpenCV renders clean, intuitive overlays directly on the frames

Challenges we ran into

Object detection, sometimes the model selects frames where nothing appears on the screen or it selects a human to track. Generating accurate vectors with an LLM is pretty hard.

Accomplishments that we're proud of

Solving a problem me and my peers have

What we learned

I learned that you need to be resilient and patient with the project. There will be lots of time where it doesn't work and you just need to stay calm and fix it. Prompt engeneering is really important, if you are able to give a great prompt then you will get great results