GemiMent

GemiMent is an AI-powered communication coach that helps users master the art of texting. Choose a scenario, set your target tone, and get real-time feedback and expert tips!

Comment

Inspiration

We’ve all stared at a blank screen, terrified to hit "send." Whether it's negotiating a raise over email, sliding into an Instagram DM, or trying to apologize without sounding defensive, digital communication is a minefield.

We realized that people don't just need a grammar checker; they need an emotional intelligence coach. We built GemiMent to cure "texting anxiety" by letting people practice high-stakes conversations in a safe, gamified, and hilarious environment.

What it does

GemiMent is an AI-powered social skills simulator.

  • The Setup: Users select a scenario (or roll the dice for a "Random Mission"), pick their platform (WhatsApp, Email etc.), choose their skill level (Beginner to Advanced), and select a target vibe (Professional, Friendly, Gen-Z, or Roast).
  • The Practice: Users type their best draft into the text box.
  • The Vibe Check: The app sends the draft to the Gemini API, which evaluates the text based on the specific social rules of the chosen platform. It returns a Score (out of 10), a creative Rank (like "Pre-K Texter" or "Social Disaster"), personalized "Coach's Tea" (feedback explaining why it failed or succeeded), and a perfect example of what to send instead.

How we built it

  • Frontend: We built a sleek, glowing, Gen-Z-inspired UI using React and Vite. We focused heavily on UX by adding quick-select suggestion chips and a "Random Mission" generator to instantly cure writer's block.
  • Backend: We set up a lightweight Express/Node.js server to securely handle our API requests.
  • The AI Brain: We utilized the Google Gemini API (gemini-2.5-flash). The real magic is in our prompt engineering. We structured the prompt to act as a dynamic persona that adjusts its strictness based on the user's selected "Skill Level," adjusts its formatting based on the "Platform," and forces the output into a strict JSON format so our frontend can reliably parse the Score, Rank, and Feedback.

Challenges we ran into

The biggest challenge was moving away from standard text generation and forcing the LLM to act as a structured evaluator. We had to carefully engineer the prompt to ensure Gemini didn't just "fix" the text, but actually returned a consistent JSON object containing a numerical score, a creative rank, and actionable feedback every single time without breaking the app.

Accomplishments that we're proud of

We are incredibly proud of the "Roast" mode and the "Daily Missions." By adding these elements, we transformed a basic utility tool into a highly engaging, sticky educational app. The UI also came out beautifully, blending modern glassmorphism with glowing neon accents.

What we learned

We learned how incredibly capable the Gemini API is at adopting specific personas. By simply tweaking the system prompt to include parameters like "Platform" and "Skill Level," Gemini was able to seamlessly switch from writing a hyper-professional corporate email to absolutely destroying a user for using a terrible pickup line on a dating app.

What's next for GemiMent

  • Progress Tracking: User accounts to track average scores over time to see real improvement.
  • Multi-Turn Roleplay: Letting the user actually converse back and forth with the "Coach" to negotiate a tricky situation step-by-step.
  • Browser Extension: Integrating GemiMent directly into text boxes on LinkedIn, WhatsApp Web, and Gmail for real-time vibe checks.
Share this project:

Updates