• 

  Home Screen

• 

  GitHub Repo

• 

  Console Log

Inspiration

I was really close with my grandpa, who passed away recently. I kept thinking how powerful it would be if AI could let me speak with—if not him—at least a close imitation. That spark of an idea led to the creation of Resona AI, a way to reconnect with voices that shaped us.

What it does

Simply upload a voice recording or record one on the spot. The AI instantly learns the voice and speaking style, enabling real-time conversations that feel authentic, emotional, and deeply personal. Whether it’s continuing a chat with a loved one, reliving moments with someone who’s passed, or finishing an unresolved conversation—Resona AI lets you step back into those emotional connections.

How we built it

We built Resona AI using Vercel for deployment, Groq for fast and efficient LLM inference, Vapi for conversational handling, and ElevenLabs for high-fidelity voice cloning. A lot of time went into testing and refining the model to get it just right.

Challenges we ran into

Vercel struggled with the complexity of the project. The concept was too nuanced for an instant setup, so we had to step in, debug manually, and guide the platform to our desired experience. Voice training was another major challenge—we initially couldn’t get accurate voice matching until we explored and integrated more advanced APIs.

Accomplishments that we're proud of

Despite the technical hurdles, we reached a working prototype that captures the essence of what we envisioned. After extensive debugging, we built a system that feels emotionally intelligent and deeply personal. Integrating high-accuracy voice matching APIs was a huge win—finally enabling the seamless interaction we were striving for.

What we learned

We learned that emotional intelligence in AI isn’t just about words—it’s about tone, timing, memory, and continuity. Building something human-centric means thinking far beyond standard chatbot logic. We also gained deep insight into voice training, multi-API orchestration, and the patience required to get truly meaningful results.

What's next for Resona AI

While Resona AI is functioning well, there's still work to do. It doesn’t yet fully understand emotion or user nuance. Our next focus is on emotional context—training the model to better detect tone, mood, and personal history to respond more empathetically. We also plan to reduce API reliance and transition toward self-training models that allow more customization and efficiency. Ultimately, we want to open a richer, more human way to reconnect, reflect, and process complex emotions.

Demo here: https://www.loom.com/share/aa3e90a849ad4cd9a6e9bc23beba7700

Built With

• claude
• elevenlabs
• github
• groq
• vapi
• vercel
• vscode