SignBridge

Top part of screen
Taking photo of sign language and translating
forming sentence sample output

Inspiration

American Sign Language (ASL) is a rich and expressive language, but learning it can be challenging without the right tools. We built SignBridge to lower that barrier, giving learners real-time, accessible feedback to help them practice and improve their signing skills.

What it does

SignBridge uses a camera feed to recognize American Sign Language hand signs and converts them into text in photos. Users can sign individual letters, build words, and convert their signing into natural English sentences using an AI translation engine.

How we built it

Python — core application logic TensorFlow / Keras — custom-trained ASL image classification model MediaPipe — real-time hand landmark detection and cropping Streamlit — interactive web-based UI with live camera input DeepSeek API — ASL gloss to natural English sentence translation

Challenges we ran into

Dataset variability — hand signs looked different across users due to lighting, skin tone, hand size, and angle, making the model difficult to generalize Preprocessing pipeline — ensuring the camera input was correctly cropped, color-corrected, and normalized to match training data Real-time performance — balancing model accuracy with low-latency predictions in a live environment

Accomplishments that we're proud of

Built a fully functional end-to-end ASL recognition pipeline from scratch Many team members completed their first or second hackathon, gaining real-world experience working under pressure in a time-sensitive environment Successfully integrated computer vision, deep learning, and an LLM API into a single cohesive application

What we learned

Data quality is everything — the model's accuracy was directly tied to the diversity and consistency of training data How to build and deploy a real-time machine learning pipeline The importance of preprocessing consistency between training and inference How to collaborate and ship a working product under tight time constraints

What's next for SignBridge

Expand vocabulary beyond the alphabet to include full words and common phrases Improve model accuracy with a larger, more diverse training dataset Support regional ASL variations and accents Explore a wearable haptic glove to give users physical feedback while learning to sign Add a learning mode with guided practice exercises and scoring