AI Digital Twin

A context-adaptive persona engine that creates a high-fidelity AI representation of a professional. Visitors interact with the digital twin through a chat interface, getting accurate responses grounded in real career data, communication style, and professional history.

Live: https://d5zo5ki02wqtq.cloudfront.net

Architecture

Browser → CloudFront (CDN) → S3 (Static Frontend)
                                    ↓
                            API Gateway (HTTP)
                                    ↓
                           AWS Lambda (FastAPI)
                                    ↓
                         AWS Bedrock (Nova Lite)
                                    ↓
                        S3 (Conversation Memory)

Tech Stack

Layer	Technology
Frontend	Next.js 16, React 19, Tailwind CSS, react-markdown
Backend	Python 3.12, FastAPI, Mangum (Lambda adapter)
AI	AWS Bedrock (Amazon Nova Lite), dynamic prompt engineering
Infrastructure	Terraform, AWS Lambda, API Gateway v2, CloudFront, S3, Route53, ACM
CI/CD	GitHub Actions with OIDC authentication

Key Features

Persona Engine

• Dynamically loads biographical facts, LinkedIn profile, communication style, and professional summary into the system prompt
• Maintains conversation context across stateless Lambda invocations via S3 JSON persistence
• Supports multi-turn conversations with a 20-message context window
• Guardrails against hallucination, jailbreak attempts, and unprofessional interactions

Chat UI

• Profile hero section with headshot and professional title
• Circular avatar headshots on all assistant messages with online status indicator
• Clickable conversation starter chips for first-time visitors
• Markdown rendering for AI responses (bold, lists, code blocks, headings)
• Multi-line textarea input with auto-resize (Enter to send, Shift+Enter for new line)
• Responsive chat height that adapts to screen size
• Animated typing indicator during AI response generation

Infrastructure

• Fully reproducible via Terraform with multi-environment support (dev/test/prod)
• Serverless compute with AWS Lambda — zero infrastructure maintenance
• Global edge delivery via CloudFront with SSL/TLS and custom domain support
• OIDC-based GitHub Actions deployment — no hardcoded AWS credentials
• API throttling (burst: 10, rate: 5) to control costs

Project Structure

├── backend/
│   ├── server.py              # FastAPI application
│   ├── lambda_handler.py      # Lambda entry point (Mangum)
│   ├── context.py             # System prompt builder
│   ├── resources.py           # Data loader (PDF, facts, style)
│   ├── deploy.py              # Lambda package builder (Docker)
│   ├── requirements.txt
│   └── data/
│       ├── facts.json         # Biographical facts
│       ├── summary.txt        # Professional summary, skills, projects
│       ├── style.txt          # Communication style guide
│       └── linkedin.pdf       # LinkedIn profile
├── frontend/
│   ├── app/
│   │   ├── page.tsx           # Home page with profile hero
│   │   ├── layout.tsx         # Root layout
│   │   └── globals.css        # Global styles
│   ├── components/
│   │   └── twin.tsx           # Chat component
│   ├── public/
│   │   └── hopeogbons.png     # Headshot avatar
│   └── package.json
├── terraform/
│   ├── main.tf                # Lambda, API GW, CloudFront, S3, Route53
│   ├── variables.tf           # Configurable inputs
│   ├── outputs.tf             # Deployment URLs
│   ├── backend.tf             # S3 state backend
│   └── versions.tf            # Provider versions
├── scripts/
│   ├── deploy.sh              # Deployment orchestration
│   └── destroy.sh             # Infrastructure teardown
└── .github/workflows/
    ├── deploy.yml             # Auto-deploy on push to main
    └── destroy.yml            # Manual teardown with confirmation

Local Development

Prerequisites

• Python 3.12
• Node.js 20+
• AWS CLI configured with Bedrock access
• Docker (for Lambda packaging only)

Backend

cd backend
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python server.py

The backend runs on http://localhost:8000. Conversations save locally to ../memory/.

Frontend

cd frontend
npm install
npm run dev

The frontend runs on http://localhost:3000 and connects to the backend automatically.

Environment Variables

No .env file is required for local development. Defaults handle everything:

Variable	Default	Description
USE_S3	false	Use local file storage instead of S3
CORS_ORIGINS	http://localhost:3000	Allowed frontend origin
BEDROCK_MODEL_ID	amazon.nova-lite-v1:0	AWS Bedrock model
DEFAULT_AWS_REGION	us-east-1	AWS region

To override, create backend/.env:

DEFAULT_AWS_REGION=us-east-2
BEDROCK_MODEL_ID=amazon.nova-micro-v1:0

Deployment

Direct Deploy (from local machine)

DEFAULT_AWS_REGION=us-east-2 ./scripts/deploy.sh dev

This builds the Lambda package, applies Terraform, builds the frontend, and syncs to S3.

GitHub Actions

Push to main triggers automatic deployment. Manual dispatch supports environment selection (dev/test/prod).

Required GitHub Secrets:

Secret	Description
AWS_ACCOUNT_ID	12-digit AWS account ID
DEFAULT_AWS_REGION	Target region (e.g., us-east-2)
AWS_ROLE_ARN	IAM OIDC role ARN for GitHub Actions

AWS Prerequisites

• S3 bucket for Terraform state: twin-terraform-state-<ACCOUNT_ID>
• DynamoDB table for state locking: twin-terraform-locks
• IAM OIDC identity provider for GitHub Actions
• Bedrock model access enabled for Nova models

Teardown

DEFAULT_AWS_REGION=us-east-2 ./scripts/destroy.sh dev

API Endpoints

Method	Path	Description
GET	/	Service info (model, storage type)
GET	/health	Health check
POST	/chat	Send a message, get AI response
GET	/conversation/{session_id}	Retrieve conversation history

Chat Request

{
  "message": "Tell me about your experience with AWS",
  "session_id": "optional-existing-session-id"
}

Chat Response

{
  "response": "I've been working extensively with AWS...",
  "session_id": "generated-or-existing-session-id"
}