PABPAT/TCI_Shield
0
TCI Shield
AI-powered trade credit insurance underwriting using Amazon Nova.
Overview
TCI Shield is a voice and text-based trade credit insurance underwriting agent. A business owner speaks to Alex, an AI underwriter powered by Amazon Nova 2 Sonic, provides their business and buyer details, and receives a policy — entirely through conversation.
Architecture
Voice (Nova 2 Sonic) Text Chat (Nova Lite)
| |
| |
+-----------+------------+
|
Strands Agent
(Nova Lite)
|
+-----------+------------+
| | |
Underwriting Document DynamoDB
Engine Extractor
(Risk Score) (Nova (Policies)
Multimodal)Nova Models Used
- Amazon Nova 2 Sonic — real-time bidirectional voice conversation
- Amazon Nova Lite — agent reasoning and underwriting orchestration
- Amazon Nova Multimodal — financial document extraction (PDF, DOCX, XLSX)
Features
- Voice-based insurance application via Amazon Nova 2 Sonic
- Text chat application via AWS Strands Agents and Nova Lite
- Automated risk scoring engine with 6 weighted dimensions
- Financial document extraction using Nova Multimodal
- Policy generation with 3 coverage options
- Policy issuance and storage in AWS DynamoDB
- Streamlit UI with voice agent, text chat, policy summary, and dashboard
Risk Scoring Model
The underwriting engine scores risk across 6 dimensions:
- Industry risk
- Buyer country risk
- Financial health ratios
- Payment terms
- Portfolio concentration
- Loss history
Scores map to 4 policy tiers: Standard, Enhanced, High Risk, or Declined.
Project Structure
Trade_Credit_Agent/
app.py # Streamlit entry point
pages/
1_Voice_Agent.py # Nova Sonic voice interface
2_Text_Chat.py # Nova Lite text chat
3_Policy_Summary.py # Latest policy view
4_Dashboard.py # All policies dashboard
tci_agent.py # Strands Agent + underwriting tools
voice_agent.py # Nova Sonic bidirectional streaming
underwriting_engine.py # Risk scoring engine
document_extractor.py # Nova Multimodal document extraction
database.py # DynamoDB operations
models.py # Pydantic validation models
config.py # AWS configuration
.env.example # Environment variables templateSetup
Prerequisites
- Python 3.13
- AWS account with Bedrock access
- AWS CLI configured with
aws configure
Installation
git clone https://github.com/PABPAT/TCI_Shield.git
cd TCI_Shield
python -m venv venv
source venv/bin/activate
pip install -r requirements.txtEnvironment Variables
Copy .env.example to .env and fill in your values:
cp .env.example .envCOMPANIES_HOUSE_API_KEY=your_key_hereAWS Setup
aws configureEnter your AWS access key, secret key, and set region to us-east-1.
Enable these models in AWS Bedrock console:
amazon.nova-lite-v1:0amazon.nova-2-sonic-v1:0
Run
streamlit run app.pyOpen http://localhost:8501 in your browser.
Dependencies
strands-agents
strands-agents-tools
boto3
aws-sdk-bedrock-runtime
smithy-aws-core
pyaudio
pydantic
streamlit
python-dotenvBuilt With
- Amazon Nova 2 Sonic
- Amazon Nova Lite
- Amazon Nova Multimodal
- AWS Strands Agents
- AWS Bedrock
- AWS DynamoDB
- Python 3.13
- Streamlit
Hackathon
Built for the Amazon Nova AI Hackathon — March 2026.
Category: Voice AI + Agentic AI
