Inspiration Losing someone you love is one of the most painful human experiences. The hardest part isn't the funeral — it's the silence after. The unanswered texts. The voice notes you replay just to hear them one more time. The conversations you never got to finish. We built GhostDoc because grief is universal. 1.4 billion people in India alone have lost someone close. Most of them would give anything for one more conversation. We wanted to give them that.

What it does GhostDoc lets you have one final conversation with someone you've lost. You upload their WhatsApp chat export, a photo, and voice notes. Our AI reads through every message, extracts their personality, speech style, common phrases, memories, and language patterns — and builds a living digital personality of that person. Then you can talk to them again, in their exact words, in their language, in their voice. The split screen interface shows their photo coming alive on the left while the conversation happens on the right — with voice responses playing automatically after every message.

How we built it We built GhostDoc using the full Amazon Nova stack via Amazon Bedrock: Amazon Nova Pro powers the personality engine — reading thousands of WhatsApp messages and extracting speech patterns, emotional tendencies, memories, and language mix to build a rich personality graph. Amazon Nova Sonic handles voice synthesis — converting AI responses into natural speech that captures the person's tone and cadence. Amazon Nova Omni processes multimodal inputs — reading text chats, understanding photos, and processing voice notes simultaneously. On the frontend we used React with Framer Motion for cinematic animations and Tailwind CSS for styling. Firebase handles authentication, Firestore stores personality graphs and conversation history, and Firebase Hosting serves the live app.

Challenges we ran into The biggest challenge was building a personality engine that felt genuinely human. Getting the AI to respond not just accurately but emotionally — in the exact language mix, with the exact phrases, referencing real memories — required extensive prompt engineering with Amazon Nova Pro. We also faced challenges with real-time voice synthesis for Hinglish (Hindi + English mixed speech), building the split-screen talking portrait interface, and handling large WhatsApp exports efficiently within API token limits.

Accomplishments that we're proud of We are incredibly proud of the emotional depth GhostDoc achieves. When we tested it with a real WhatsApp export, the AI responded with "Beta, Bhagwan ka shukar hai ki tum mujhse baat kar rahe ho" — exactly how a real Indian grandmother would speak. It gave us chills. We're also proud of the cinematic UI — the dark spooky theme, the pulsing portrait with animated rings, the waveform that comes alive when the person speaks. It feels like a sacred space, not just an app.

What we learned We learned that AI can do more than automate tasks — it can heal. The most powerful applications of AI are the ones that address the deepest human emotions. We learned that personality is not just about words — it's about language mix, emotional patterns, specific memories, and the way someone makes you feel. Amazon Nova Pro surprised us with how deeply it could understand and replicate human personality from just chat history.

What's next for GhostDoc The vision for GhostDoc goes far beyond a hackathon project:

Voice cloning — using actual voice recordings to recreate their exact voice Video portraits — lip-synced video responses using their real photos Memory timeline — a visual journey through their life built from their chats and photos Multi-language support — Tamil, Telugu, Marathi, Bengali and more Indian languages Mobile app — iOS and Android for accessibility Memorial pages — shareable public tributes that family members can contribute to Grief counselor integration — partnering with mental health professionals to use GhostDoc as a therapeutic tool

GhostDoc is not just an app. It is a sacred space where memories live forever.

Built With

  • amazon-bedrock
  • amazon-nova-omni
  • amazon-nova-pro
  • amazon-nova-sonic
  • elevenlabs
  • firebase-authentication
  • firebase-firestore
  • firebase-hosting
  • framer-motion
  • groq-llama-3.3-70b
  • javascript
  • react
  • tailwind-css
Share this project:

Updates