Broski - AI Conversation Coach

Note: Due to Wi-Fi issues at the venue, we couldn’t demonstrate the Hume integration in the demo. Hume analyzes vocal tone to detect emotions such as confidence, hesitation, and nervousness, then provides feedback based on these signals. Incorporating Hume allows our system to go beyond text-based interaction, offering how users sound, not just what they say.

Inspiration

We’ve all been there, heart racing before an interview, voice shaking during a presentation, or overthinking what to say in a social setting. Practicing alone rarely gives the insights we need. Broski was born from that shared struggle. We wanted a tool that could listen, respond, and coach us in real time, a way to practice confidently and get the kind of feedback a real mentor or audience would give. So we built Broski to make self-practice feel human.

What it does

Broski offers three modes designed for real-life scenarios:

Professional Mode: Mock interviews with an AI companion that asks questions in both voice and text, provides and adaptive voice and text follow-ups
Social Mode: Freestyle conversation practice for confidence and flow.
Presentation Mode: Speak freely and receive deep feedback on clarity, tone, grammar, and cohesion.

After each session, users receive a comprehensive analysis that includes a video playback, an AI-generated transcript with paragraph-level (per-question) feedback on both delivery and content, as well as an overall assessment of grammar, clarity, cohesion, and suggested improvements. It’s like having a personal coach who listens, analyzes, and helps you grow stronger, one conversation at a time.

How we built it

Frontend: A React dashboard inspired by the simplicity of Google Meet, built for ease and familiarity. Backend: Flask and SQLAlchemy manage sessions, question generation, and transcripts. AI Integration: Groq for fast speech-to-text. Claude for generating adaptive questions and in-depth analysis. Hume for emotion and tone evaluation. CodeRabbit for our product development.

We connected these systems through API endpoints that handle real-time updates, async feedback, and end-of-session reporting. It will create a new experience from speaking to self-reflection.