BabelAI

Inspiration

Our team here at BabelAI noticed a limitation with Google Translate: the lack of clarity in tone. Your tone will change depending on whether you are talking to an esteemed CEO, or to your favorite toddler cousin. To remedy this, we built this application to provide users with a variety of options when it comes to clarifying one’s conversational tone.

What it does

The user starts by allowing their conversation partner to speak in their language, which BabelAI will translate into the user’s language. Then, the user can draft a basic idea of how they want to respond, and the app will give three responses that vary in tone. The user has the freedom to choose which response they feels captures their intent the best, and the conversation can flow from there!

How we built it

It was important that our translation services were as accessible as possible, so we decided to make a mobile application. We decided that the best framework was Flutter due to its cross-platform capabilities, which allows it to be run on both iOS and Android devices. Regarding APIs, we utilized Google Translate and GeminiAI for translation and response suggestion, respectively.

Challenges we ran into

Our team wanted to ensure that Babel had unique functionality when compared to an app like Google Translate, and so our ambitions were originally to create dynamic and personal profiles for users’ conversation partners. However, due to time and technology limitations, we decided to focus on the core functionality of tone recognition and recommendation.

Additionally, we found challenges with the Gemini LLM, as we needed to feed it highly specific prompts in order to get the desired output with the correct tone and translation. When fed too much instructions, the Gemini LLM tends to sputter out, which meant we had to modify our prompt engineering process. We overcame this challenge by streamlining the prompts, making them more clear and concise for Gemini, which allowed it to yield much better responses.

Accomplishments that we're proud of

We are pleased with our utilization of the Gemini LLM to provide users with conversational freedom and specificity. This sets Babel apart from more serialized translation apps like Google Translate by utilizing the powerful AI LLM technology alongside phrase for phrase translation, giving users much more meaningful exchanges and interactions that break down language barriers.

What we learned

• API Integration: We mastered making API calls using Flutter's built-in libraries, particularly for Google Translate, GeminiAI, and Speech-to-Text. This enhanced our app's functionality and data access.
• Prompt Engineering for LLM: We refined our understanding of Language Model (LLM) outputs through prompt engineering. Specifically, we learned that LLM models, such as GeminiAI, perform better with specific, yet succinct prompts.
• Strategic Framework Installation: We discovered the importance of installing frameworks and libraries early in the development process, significantly saving time and streamlining the workflow.

What's next for BabelAI

1. Profiles for Conversationalists: We will introduce profiles for conversationalists, including information such as name, relationship to the user, and past conversation history. This will facilitate smoother and more personalized interactions.
2. Deeper Training for LLM: To achieve a more consistent tone from the user, we will provide deeper training for the Language Model (LLM). By leveraging past conversations and analyzing the ways users chose to respond given the options provided, BabelAI will be better equipped to understand and reflect the user's tone accurately.
3. User Authentication: Implementing user authentication will allow us to store profiles, conversation history, and relationships securely. This will enhance user experience and ensure data privacy.

Built With

• dart
• flutter
• gemini
• google-translate
• xcode