Translation made easy!
Function
A live multi-modal translation agent based on Gemini 2.5 Flash Native Audio. Live agent translates visually with text and audibly with voice. It works from target language to base language and base to target, with a "Snap to Translate" feature to help with everyday translations like menus, signs, and stores. Intended to be a mobile app with a single push to start agent button for accessibility in mind. Currently supports English, Spanish, French, German, Japanese, Chinese, Hindi, Arabic.
Tools
I used the Google GenAI SDK in AI Studio Build to code the entire app with Gemini 3.1 Pro Preview. Multi-turn prompting and app iteration. It isn't perfect. The generated text that appears over the live video is in markdown and shows some thinking. I hope to fix that after the challenge. The agent stopped working every time I tried. Video and PDF were generated in NotebookLM using my resources, pics, and links.
Challenges
I initially started in Antigravity, attempting to integrate the new WAXAL dataset into a live agent using Firebase Skills. After having some issues and being unhappy with my build, I changed gears days before the deadline. It has been a fairly chaotic build juggling solo dad life and traveling, but it did offer a creative way to stress test my app, as seen in the video.
This app was meant to be deployed via AI Studio. The publish feature is currently under maintenance, so I used Vercel, AI Studio, Github, and GCP to deploy the app. This was my first time using Vercel, so I will continue to work on the deployment to ensure a working app.
Vercel App Link
Changes
I learned a lot from this challenge. One is to not wait until the night before the deadline to submit. More than that, I need to have more faith in myself. Doubting my ability and skill has prevented me from entering things like this is the past. I'm proud that I produced a somewhat working app!
What's next for TranslateRT
Definitely fix the live generated text. The snap to translate works just fine, but audio translations need work. It needs more testing and support for other languages. Snapped images can already be saved, but I'd like to have a way to save translations as a txt doc. I even considered buying a cheap domain for it.
Built With
- aistudio
- cloudrun
- gemini
- html
- typescript
- vercel
Log in or sign up for Devpost to join the conversation.