Automated Client Engagement Platform

An AI-powered platform that transforms traditional cold calling into data-driven, personalized client engagement for financial institutions.

Problem

Financial institutions face significant challenges with traditional outreach:

• Low Success Rates - Cold calls achieve only 10-15% effectiveness
• Time-Intensive Research - Manual client research requires substantial resources
• Missed Opportunities - Without data-driven insights, clients are contacted at suboptimal times
• Scalability Constraints - Human agents cannot efficiently analyze large volumes of client data

Solution

A four-stage automated pipeline that analyzes customer data, generates personalized strategies, and executes outreach through AI-powered voice calls.

┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│  1. DATA        │ →  │  2. PROFILE     │ →  │  3. OUTREACH    │ →  │  4. INSIGHTS    │
│  INGESTION      │    │  GENERATION     │    │  EXECUTION      │    │  DASHBOARD      │
└─────────────────┘    └─────────────────┘    └─────────────────┘    └─────────────────┘
        │                      │                      │                      │
        ▼                      ▼                      ▼                      ▼
   Customer DB            AI Agents             ElevenLabs              Track all
   (PostgreSQL)           analyze &             Voice Calls            interactions
                          strategize            + Email/SMS            & outcomes

Key Innovation

Distributed AI Architecture - Using smaller, specialized models for dedicated tasks rather than one large model. Achieved 30% improvement in efficiency.

Built With

Backend

• Python 3.9+ - Core language
• FastAPI - REST API framework
• SQLAlchemy - ORM for database operations
• PostgreSQL - Database

Frontend

• React 18 - UI framework
• Vite - Build tool
• Tailwind CSS - Styling
• React Router - Navigation

AI/ML Services

• Google Gemini (via OpenRouter) - Strategy generation & transcript analysis
• ElevenLabs - AI voice synthesis for outbound calls

Communication

• Twilio - Outbound phone calls & SMS

Project Structure

cxc_hackathon/
├── api/                         # FastAPI backend
│   └── main.py                  # REST API endpoints
│
├── DB/                          # Database layer
│   ├── db_manager.py            # SQLAlchemy models & CRUD
│   ├── customers/               # 10 sample customer profiles (JSON)
│   └── SCHEMA.md                # Database schema docs
│
├── strategy_planning/           # AI strategy generation
│   ├── prompt_template.py       # Prompt builder with placeholders
│   ├── strategy_pipeline.py     # Main strategy pipeline
│   └── gemini_example.py        # Gemini API integration
│
├── transcript_analysis/         # Call transcript processing
│   ├── transcript_analyzer.py   # AI-powered transcript analysis
│   └── prompt_template.py       # Analysis prompts
│
├── prompts/                     # Master AI prompts
│   ├── master_prompt_voice.md   # Voice call agent prompt
│   └── master_prompt_email.md   # Email/SMS prompt
│
├── tts-stt/                         # Text-to-speech
│   └── text_to_speech.py        # ElevenLabs integration
│
├── web/                         # React dashboard
│   └── src/
│       └── pages/
│           ├── CustomerList.jsx
│           ├── CustomerDetail.jsx
│           └── CallHistory.jsx
│
└── tests/outbound_calling/      # Twilio + OpenAI prototype
    ├── realtime_server.py       # WebSocket bridge
    └── server.py                # Flask server for TwiML

Customer Profile Types

The system classifies customers into 5 profile types for personalized engagement:

Type	Name	Credit Score	Days Past Due	Approach
1	Low-Risk Recovery	700+	0-30	Friendly, fee waiver offers
2	Early Financial Stress	650+	15-60	Educational, helpful
3	Moderate Hardship	580-650	60-120	Payment plan options
4	Severe Crisis	<580	120+	Compassionate, resources
5	High-Value VIP	Any	Any	Premium treatment

Quick Start

Prerequisites

• Python 3.9+
• Node.js 18+ (for frontend)
• uv package manager

1. Clone and Setup

git clone https://github.com/dklpp/cxc_hackathon.git
cd cxc_hackathon

# Install uv (if not installed)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install Python dependencies
uv sync --no-install-project

# Activate virtual environment
source .venv/bin/activate

2. Configure Environment

cp .env.example .env

Edit .env with your API keys:

TWILIO_ACCOUNT_SID=your_twilio_sid
TWILIO_AUTH_TOKEN=your_twilio_token
ELEVEN_LABS_API_KEY=your_elevenlabs_key

3. Initialize Database

uv run python DB/db_usage_example.py

This creates the SQLite database and loads 10 sample customers.

4. Run the API Server

uv run uvicorn api.main:app --reload

API available at http://localhost:8000

5. Run the Frontend (optional)

cd web
npm install
npm run dev

Dashboard available at http://localhost:5173

Usage Examples

Generate a Strategy for a Customer

from strategy_planning.prompt_template import build_strategy_prompt

prompt = build_strategy_prompt(
    customer_first_name="Maria",
    customer_last_name="Santos",
    credit_score=785,
    days_past_due=15,
    preferred_channel="call"
)
# Send to Gemini for strategy generation

Analyze a Call Transcript

from transcript_analysis import TranscriptAnalyzer

analyzer = TranscriptAnalyzer()
result = analyzer.analyze_transcript_file(
    "call_recording.txt",
    customer_id=1
)
print(result['call_outcome'])
print(result['recommendations'])

Sample Customers

The system includes 10 realistic customer profiles:

1. Maria Santos - Good standing, forgot payment while traveling (15 days)
2. David Kim - Autopay card expired (5 days)
3. James Wilson - Laid off from GM (62 days)
4. Tyrone Washington - Medical emergency (58 days)
5. Robert Miller - Job loss + divorce (328 days)
6. Sarah Chen - New job starting, needs 2-week grace (8 days)
7. Patricia Nguyen - Bank processing delay (3 days)
8. Michael O'Brien - Self-employed, slow season (95 days)
9. Linda Martinez - Recently widowed (35 days)
10. Emmanuel Okafor - Recent grad, first missed payment (12 days)

API Endpoints

Method	Endpoint	Description
GET	/customers	List all customers
GET	/customers/{id}	Get customer details
GET	/customers/{id}/summary	Get customer with debts & history
POST	/customers/{id}/call	Initiate AI call to customer
GET	/communications	List all communication logs

Expected Outcomes

• Increase engagement success rates beyond industry standard 10-15%
• Reduce research and preparation time per client interaction
• Scale personalized outreach without proportional increases in resources
• Improve customer satisfaction through relevant, well-timed communications