Eternal Connect

Inspiration

Losing someone you love leaves a silence that’s hard to fill. We all carry questions, memories, and words left unsaid. Eternal Connect was born from that longing, the wish to have one more conversation, to hear their voice again, to feel close even when they’re gone. We asked ourselves, what if technology could help bridge that emotional gap in a respectful and meaningful way?

What it does

Eternal Connect lets people have AI-powered conversations with digital versions of their lost loved ones. By using memories, voice recordings, and shared stories, we create a custom conversational model that sounds and speaks like the person you miss. It’s not about replacing them, it’s about remembering them, honoring their voice, and finding comfort in connection.

How we built it

Eternal Connect is built on a robust full-stack architecture. The frontend is developed with React.js and TypeScript, styled using Tailwind CSS and Shadcn UI components for a modern and responsive user experience. We use Supabase as our backend, providing a PostgreSQL database for storing memorial data, user profiles, memories, and media files, along with Supabase Auth for secure user management and Supabase Storage for handling uploaded avatars and media.

The core of our AI-powered conversations runs on Supabase Edge Functions, which orchestrate interactions with various AI services:

Groq AI (specifically the llama-3.3-70b-versatile model) powers the intelligent text responses, ensuring conversations are snappy, context-aware, and reflect the loved one's personality and life story. ElevenLabs (using the eleven_flash_v2_5 model) is integrated for realistic voice cloning and text-to-speech, allowing users to hear responses in the familiar voice of their loved one. Tavus is utilized for creating and managing AI video replicas and personas, enabling lifelike video conversations that capture natural expressions and emotional depth. We've meticulously crafted system prompts and integrated comprehensive biographical and personality data to ensure the AI models generate responses that are deeply personalized and emotionally intelligent, making each interaction feel authentic.

Challenges we ran into

Emotion is messy and deeply human. One of the biggest challenges was balancing the technical with the ethical. We had to be careful about consent, memory sensitivity, and emotional safety. Technically, integrating multiple advanced AI APIs (Groq, ElevenLabs, Tavus) and ensuring seamless data flow and context management across them was complex. Specifically, creating and managing persistent AI personas with rich story context in Tavus, and ensuring the Groq model accurately reflected the nuanced personality and life details, required significant fine-tuning of prompts and parameters. Keeping the conversations from sounding robotic or off-putting took a lot of iteration on temperature settings and prompt engineering.

Accomplishments that we're proud of

We built something that can make people cry, in a good way. We're proud of how we handled sensitive data, respected boundaries, and created a space that feels gentle and human. We're particularly proud of the seamless integration of multiple AI services to deliver a truly multi-modal conversational experience (text, voice, video) that feels deeply personal and emotionally resonant. The ability to automatically create and leverage comprehensive AI personas based on the memorial's story context is a significant technical achievement.

What we learned

We learned that AI can be more than just a tool, it can be a bridge. We saw firsthand how voice, tone, and memory shape how we grieve and remember. We also learned that trust is everything. People need to know their memories are safe. Building with care and transparency made all the difference.

What's next for Eternal Connect

We're exploring partnerships with grief counseling services, adding more emotional safeguards, and continuously improving memory accuracy and AI persona fidelity over time. We plan to integrate more advanced emotional nuances into voice responses, potentially leveraging future ElevenLabs models like Eleven v3. We also want to create guided experiences for significant dates like birthdays, anniversaries, or specific life advice sessions, allowing users to reconnect in even more meaningful ways. Ultimately, we hope Eternal Connect becomes a compassionate companion in healing, not just a tool for conversation, recognizing that everybody deals with grief in their own unique ways.