Skip to content

helloKittyCatSweet/BeWithMe

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

🎙️ BeWithMe - AI Emotional Companion & Memory Persistence System

Digital Platform for Emotional Assets Powered by Mistral Large 2, ElevenLabs & Blockchain

Built for the NVIDIA Mistral Worldwide Hackathon 2026. BeWithMe is dedicated to creating meaningful conversations with loved ones through AI & Web3, ensuring these precious memories are eternal and owned by you.


� Demo

1. Login & Security Authentication

Secure login portal ensuring privacy and data protection. Login Interface

2. Create Digital Loved One (Create Agent)

Easily configure personality traits, upload photos, and initiate a digital life. Create Agent

3. Voice Cloning

Upload a 30-second audio clip to replicate a loved one's voice, supporting various emotional expressions. Voice Cloning

4. Immersive Phone Call Experience

Core Feature: Realistic call interface supporting real-time voice conversations, delivering emotional comfort as if they were by your side. Phone Call Interface

5. Emotional Asset Ownership & Verification

Leveraging blockchain technology for ownership verification, ensuring every conversation and memory is traceable and immutable. Blockchain Verification

6. Trust & Security Architecture

Multi-layered security framework protecting your emotional data. Trust & Security


�📖 Project Intro: What is BeWithMe?

BeWithMe is a full-stack AI emotional companion system designed to "replicate" the digital persona of loved ones. It's more than just a chatbot; it's a closed-loop system integrating voice, personality, memory, and ownership.

  • Core Pain Point: Addressing the longing for lost or distant loved ones by achieving "Digital Immortality" through technology.
  • Three Key Stages:
    1. Replication: 30-second voice cloning and personality modeling via character descriptions.
    2. Evolution: Fine-tuning Mistral models using Google Colab's powerful compute to make the "soul" more authentic.
    3. Persistence: Leveraging blockchain for decentralized storage, ensuring emotional data survives beyond any single service provider.

🏗️ System Architecture

BeWithMe Architecture

Figure 1: BeWithMe System Technical Architecture. Illustrates the end-to-end flow from frontend interaction to the Mistral backend engine, Google Colab fine-tuning, and blockchain-based ownership verification.

graph TD
    subgraph Frontend [User Interface - Streamlit]
        UI[Interactive UI]
        AudioRec[Voice Recording / Config Upload]
    end

    subgraph Backend [Logic Center - FastAPI]
        Brain[Mistral Large 2 - Core Conversation Engine]
        TTS[ElevenLabs - Voice Cloning / Synthesis]
        Manager[Config Manager]
    end

    subgraph Training [Evolution & Fine-tuning - External]
        Colab[Google Colab - Model Fine-tuning]
        WandB[W&B - Training Monitoring]
        Weave[W&B Weave - Model Comparison & Eval]
    end

    subgraph Web3 [Emotional Persistence - Blockchain]
        Contract[Solidity - MemoryLock Contract]
        Chain[Ethereum Sepolia]
    end

    UI --> AudioRec
    AudioRec --> Brain
    Brain --> TTS
    TTS --> UI

    Brain -.-> Weave
    Colab --> WandB
    WandB -.-> Brain
    
    Brain -.-> Contract
    Contract --> Chain
Loading

🛠️ Tech Stack

Module Technology Role
LLM (Brain) Mistral Large 2 Core conversation engine for empathy and logic
Voice (Voice) ElevenLabs API 30s instant cloning for high-fidelity voice replies
Compute Google Colab High-performance GPU power for Model Fine-tuning
Monitoring W&B (Weights & Biases) Real-time tracking of fine-tuning metrics
Evaluation W&B Weave Comparison of model versions and performance evaluation
Backend FastAPI High-performance async Python framework for business flow
Frontend Vue.js 3 + TypeScript Modern responsive Web UI for an immersive user experience
Web3 (Ownership) Solidity + Sepolia Smart contracts managing IPFS hashes for asset ownership

✨ Core Highlights

1. Deep Fine-Tuning & W&B Evaluation (Fine-Tuning & Eval)

We leverage Google Colab for deep fine-tuning of Mistral models. Metrics are monitored via W&B, and we use W&B Weave for multi-dimensional evaluation:

  • Latency: Optimizing response speed for real-time call experiences.
  • Empathy: Evaluating the AI's ability to provide emotional comfort.
  • Human Likeness: Ensuring tone and pace match the specific loved one.

W&B Model Comparison

2. Blockchain Persistence (MemoryLock)

Solving the "Ownership of Emotional Assets" problem via blockchain. By anchoring material hashes and model weights on Sepolia:

  • Digital Ownership: Only family members with specific private keys can "awaken" the persona.
  • Eternal Storage: Decentralized persistence ensures memories aren't tied to a single provider.

3. Ethics & Safety 🛡️

  • Verification: Integrated kinship verification to prevent misuse of voice cloning technology.
  • Data Privacy: Sensitive audio is used strictly for model training with user consent.
  • Psychological Guidance: AI provides timely emotional support and reminders within conversations.

🌈 Future Vision

  • Multimodal Memories: Expanding from voice to Digital Humans and immersive VR environments.
  • Digital Legacy DAO: Establishing decentralized organizations to maintain a family's digital assets.
  • Cross-Era Dialogues: Allowing wisdom and love to transcend time as a part of family heritage.

🚀 Quick Start

Launch Application

chmod +x scripts/run_all.sh
./scripts/run_all.sh

Access points:

  • 🎨 Frontend UI: http://localhost:8501
  • 🔧 API Docs: http://localhost:8000/docs

🏆 Hackathon Judges: See DEMO_GUIDE.md for the evaluation walkthrough.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors