Inspiration

The inspiration for ToneLift AI came from my experiences in sales and professional communication. I frequently encountered situations where crafting the right message was crucial - whether negotiating with clients, providing feedback to team members, or handling sensitive discussions. I noticed that it often took considerable time to compose messages that maintained effectiveness while conveying empathy and professionalism. A key observation was that while AI could theoretically generate these messages from scratch, this wouldn't help professionals develop their communication skills. Instead, I envisioned a tool that would work collaboratively with users, helping them understand how to improve their communication while preserving their authentic voice.

What it does

ToneLift AI transforms messages to achieve better communication outcomes while maintaining the core message. It specializes in business contexts like:

  • Negotiating with clients and stakeholders
  • Providing constructive feedback to team members
  • Delivering difficult decisions or news to employees
  • Handling internal conflicts professionally

While also supporting broader applications like:

  • Filing effective customer complaints
  • Navigating sensitive family discussions

How we built it

  • Frontend: Next.js, Typescript and Tailwind CSS
  • AI functionalities:
    • Rewriter API for message transformation
    • Prompt API for improvement analysis
    • Gemini Nano for enabling on-device natural language processing and advanced text rewriting

The architecture focuses on:

  • Local processing using Chrome's built-in AI capabilities
  • Real-time feedback and analysis
  • Clean, intuitive user interface
  • Voice input integration for accessibility

Challenges we ran into

  1. Limited Response Sophistication I found significant differences in response quality between Chrome's Built-in AI and more advanced models like meta/llama-3.1-405b-instruct. Here's a direct comparison using a business negotiation scenario:
  • Input Message: Hi Henry, these changes weren't part of the deal, and we can't just add them without more time and money. Let me know when you want to talk about it.

  • Chrome Built-in AI Response: Hi Henry, these changes are not part of the agreement, and we can't simply add them without additional time and resources. Please let me know when you're available to discuss.

->Analysis focused on surface-level improvements:

  • Basic sentence structure changes
  • Simple word replacements (e.g., "deal" → "agreement")
  • Standard polite closing Level of improvements did not drastically change even when I changed the temperature from 0.1 to 0.9.

  • Meta/Llama-3.1-405b Response: Hi Henry, I understand that changes can be necessary, but I also want to ensure we can deliver the best results within our agreed-upon timeframe and budget. The additional changes you've requested are significant, and I'd like to discuss how we can accommodate them while meeting your needs. Can we schedule a call to explore options and find a solution that works for everyone?

-> Analysis showed deeper improvements:

  • Added empathetic acknowledgment of client's perspective
  • Reframed the issue to focus on solutions
  • Maintained professionalism while building relationship
  • Incorporated collaborative language
  • Used positive framing for negotiation

This comparison reveals Chrome's Built-in AI focuses primarily on basic language refinements rather than strategic communication improvements.

  1. Language Limitations
  2. When attempting to use non-English languages (Spanish and Japanese), the Rewriter API returned errors indicating these were "untested languages"
  3. Current testing has only confirmed English functionality
  4. Further testing would be needed to determine the full extent of language support

  5. Encountered unexpected behavior:

  6. The AI attempted to access external websites even when no links were provided

  7. Failed to handle inappropriate language gracefully, producing unusable responses instead of appropriate filtering

Accomplishments that we're proud of

  • Successfully implemented Chrome's Built-in AI APIs for real-time message transformation while maintaining privacy through local processing
  • Created an intuitive interface that helps users improve their communication skills rather than just generating messages
  • Implemented voice input functionality for easier message input

What we learned

  • Local processing has trade-offs between speed and sophistication
  • Importance of proper prompt engineering for consistent results

What's next for ToneLift AI

  • Integration with Rewrite API's multilingual capabilities once available
  • Chrome Extension for inline editing in email clients and messaging platforms

Built With

Share this project:

Updates