CoolFace
Apppublic

Anvesh92/CredGuard-voice-agent

sourceHugging Faceupdated 9mo agoView on Hugging Face
0likes
App README

🛡️ CredGuard: Autonomous Voice-AI Recovery Agent

CredGuard is a next-generation AI agent designed to reimagine debt recovery in the Indian Fintech landscape. By combining Voice Intelligence, Predictive Risk Modeling, and Agentic Reasoning, it transforms a traditionally friction-heavy process into a borrower-friendly, empathetic, and 100% compliant recovery journey.

🚀 Live Demo

[Interact with the Live Agent Here](https://huggingface.co/spaces/Anvesh92/CredGuard-voice-agent)

🌟 Key Features

1. Voice-First Intelligence 🎙️

  • —Real-time STT: Utilizes OpenAI Whisper-v3 (via Groq) for lightning-fast, multi-lingual speech transcription.
  • —Natural Responses: Integrated with gTTS to provide spoken feedback, ensuring an accessible experience for all borrowers.

2. Predictive Risk Assessment 📉

  • —Machine Learning Core: Features a Random Forest Classifier trained on synthetic financial datasets to predict borrower default risk in real-time.
  • —Dynamic Negotiation: The agent autonomously adjusts its tone and payment offers (e.g., suggesting EMI deferrals) based on the calculated Risk Status.

3. Agentic Reasoning & RAG Compliance ⚖️

  • —Policy Grounding: Uses a RAG (Retrieval-Augmented Generation) framework to ensure all agent interactions strictly adhere to RBI Fair Practice Codes.
  • —Autonomous Workflows: Capable of multi-step task planning—from identifying financial hardship to executing payment restructuring plans.

4. Production-Grade MLOps 🐳

  • —Containerization: Fully architected with Docker for consistent, scalable deployment.
  • —Latency Optimized: Leveraging Groq LPU™ Inference Engine to achieve sub-second response times for a production-ready feel.

🛠️ Technical Stack

CategoryTools & Technologies
LLM & ReasoningLlama-3 (8B-Instant), LangChain, Groq API
Voice IntelligenceOpenAI Whisper-v3, gTTS
Machine LearningScikit-Learn, Pandas, NumPy, joblib
Vector DatabaseFAISS (for RAG Compliance)
InterfaceGradio
InfrastructureDocker, Hugging Face Spaces

🏗️ System Architecture

The system follows a modular architecture:

  1. 1.Input Layer: Captures audio/text from the borrower.
  2. 2.Predictive Layer: Random Forest model assesses risk based on loan amount and DPD (Days Past Due).
  3. 3.Knowledge Layer: RAG retrieves relevant RBI compliance guidelines.
  4. 4.Reasoning Layer: Llama-3 synthesizes the risk score, compliance data, and user intent to generate a friendly response.
  5. 5.Output Layer: Text and Voice feedback delivered to the user.

💻 Local Installation

To run this project locally using Docker:

bash
# 1. Clone the repository
git clone [https://huggingface.co/spaces/Anvesh92/CredGuard-voice-agent](https://huggingface.co/spaces/Anvesh92/CredGuard-voice-agent)

# 2. Navigate to the folder
cd CredGuard-voice-agent

# 3. Build the Docker image
docker build -t credguard-agent .

# 4. Run the container
docker run -p 7860:7860 -e GROQ_API_KEY="your_api_key" credguard-agent

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference