MambaMentor

About MambaMentor

MambaMentor is a conversation assistant designed to elevate dialogue by offering real-time, context-aware suggestions. Inspired by Kobe Bryant’s legendary work ethic—the "Black Mamba"—and the concept of a reliable wingman, MambaMentor empowers users to navigate social and professional interactions with confidence and authenticity.

Inspiration

Our journey began with the realization that effective communication is a game changer in both personal and professional spheres.

What It Does

MambaMentor provides real-time suggestions that help users respond with clarity and creativity. These suggestions are automatically updated when new messages are sent or received and when the user clicks on the refresh button. The suggestions are steered by:

The conversation history
User-specified conversation goal
An optional persona guide
Customizable tone settings

By using these parameters to analyze ongoing dialogue, our pipeline passes them through a large language model and streams three suggestions to the user.

How We Built It

Technology Stack: Built with a Flask backend integrated with Socket.IO for real-time updates and the OpenAI API for generating suggestions.
Model Selection: After experimenting with various options, we settled on the GPT-4o model for its performance and cost.
Real-Time Processing: Our system aggregates incoming messages into a comprehensive conversation history—including an optional user-specified goal and selected persona and tone options—sends this combined context to the AI model, and instantly streams the returned suggestions to the user.
Iterative Design: We refined our prompts and interface continuously to balance technical performance with a natural, engaging experience. We also used whisperx to scrape transcripts of celebrities to create our personas and used it to train our models.

Challenges We Ran Into

iMessages Integration: Our initial goal to integrate natively with iMessages was unfeasible within our time constraints.
User Experience: Merging functionality with a polished, professional interface for a diverse audience proved challenging.
Real-Time Performance: Delivering suggestions promptly without interrupting conversation flow was a significant technical hurdle.
Model Compatibility: Transitioning to o1-mini required adjusting API calls and prompt formats due to differences in supported parameters.
Prompt Engineering: Crafting prompts that yield natural, context-aware responses required multiple iterations.

Accomplishments That We're Proud Of

Seamless Integration: Deployed real-time, streaming AI suggestions in a live chat environment, significantly boosting user engagement.
Enhanced Personalization: Incorporate user-specific persona profiles and conversation history to tailor suggestions.
Robust Prompt Design: Developed refined prompts that yield authentic, professional responses, setting us apart from generic assistants.
Scalability: Designed a modular architecture that is ready for future AI advancements.

What We Learned

Context is Key: Well-crafted prompts and rich context are crucial for high-quality AI responses.
Embrace Trade-Offs: Adapting to model limitations reinforced the importance of flexibility and creative problem-solving.

What's Next for MambaMentor

Broader Integration: Expand compatibility with popular communication platforms and enterprise solutions.
Advanced Analytics: Introduce analytics to help users understand and improve their communication patterns.
Global Outreach: Scale the platform for international markets.
iMessage Integration: Work towards seamless integration with services like iMessage, WhatsApp, and Signal.
Improved Personas: Expand persona options and further fine-tune our model for varied conversational styles.

MambaMentor is more than just an app—it’s a step towards transforming everyday personal and business interactions into more meaningful, high-impact connections.