Deploying an AI Traffic Congestion Predictor using AWS Bedrock: A Complete Overview

We all love traffic, right? The only time where I think about how I totally messed up my presentation (overthinking is a pain).

All jokes aside, I've wanted to create a project where I can look for traffic in real-time as a PoC so that I further enhance it in the future. Meet the traffic congestion predictor.

I'll walk through deploying the Traffic Congestion Predictor using AWS Bedrock. AWS Bedrock provides a fully managed service for foundation models, making it perfect for deploying AI applications. We'll cover everything from initial setup to final deployment and testing.

Now, the prerequisites

• AWS Account with appropriate permissions (Had to use my debit card for verification because I assumed it was free to use for a certain limit. Pain).
• Python 3.8+
• Traffic Congestion Predictor code (from previous development)
• AWS CLI installed and configured
• Basic knowledge of Python and AWS services will do just fine.

Step 1: Preparing Your Environment

First, set up your development environment:

# Create a new virtual environment
python -m venv bedrock-env
source bedrock-env/bin/activate  # On Windows use: bedrock-env\Scripts\activate

# Install required packages
pip install boto3 pandas numpy scikit-learn streamlit plotly

Step 2: AWS Bedrock Setup

1. Navigate to AWS Console and enable AWS Bedrock

2. Create a new model in Bedrock:

• Go to the AWS Bedrock console
• Select "Model access"
• Request access to Claude model family
• Wait for approval (usually instant but anything can happen)

Step 3: Modify Code for Bedrock Integration

Create a new file "bedrock_integration.py":

import boto3
import json
import numpy as np
import pandas as pd
from typing import Dict, Any

class TrafficPredictor:
    def __init__(self):
        self.bedrock = boto3.client(
            service_name='bedrock-runtime',
            region_name='us-east-1'  # Change to your region
        )

    def prepare_features(self, input_data: Dict[str, Any]) -> pd.DataFrame:
        # Convert input data to model features
        hour = input_data['hour']
        day = input_data['day']

        features = pd.DataFrame({
            'hour_sin': [np.sin(2 * np.pi * hour/24)],
            'hour_cos': [np.cos(2 * np.pi * hour/24)],
            'day_sin': [np.sin(2 * np.pi * day/7)],
            'day_cos': [np.cos(2 * np.pi * day/7)],
            'temperature': [input_data['temperature']],
            'precipitation': [input_data['precipitation']],
            'special_event': [input_data['special_event']],
            'road_work': [input_data['road_work']],
            'vehicle_count': [input_data['vehicle_count']]
        })
        return features

    def predict(self, input_data: Dict[str, Any]) -> float:
        features = self.prepare_features(input_data)

        # Prepare prompt for Claude
        prompt = f"""
        Based on the following traffic conditions, predict the congestion level (0-10):
        - Time: {input_data['hour']}:00
        - Day of week: {input_data['day']}
        - Temperature: {input_data['temperature']}°C
        - Precipitation: {input_data['precipitation']}mm
        - Special event: {'Yes' if input_data['special_event'] else 'No'}
        - Road work: {'Yes' if input_data['road_work'] else 'No'}
        - Vehicle count: {input_data['vehicle_count']}

        Return only the numerical prediction.
        """

        # Call Bedrock
        response = self.bedrock.invoke_model(
            modelId='anthropic.claude-v2',
            body=json.dumps({
                "prompt": prompt,
                "max_tokens": 10,
                "temperature": 0
            })
        )

        # Parse response
        response_body = json.loads(response['body'].read())
        prediction = float(response_body['completion'].strip())

        return np.clip(prediction, 0, 10)

Step 4: Create FastAPI Backend

Create "api.py:"

from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from bedrock_integration import TrafficPredictor
from typing import Dict, Any

app = FastAPI()
predictor = TrafficPredictor()

class PredictionInput(BaseModel):
    hour: int
    day: int
    temperature: float
    precipitation: float
    special_event: bool
    road_work: bool
    vehicle_count: int

@app.post("/predict")
async def predict_traffic(input_data: PredictionInput) -> Dict[str, float]:
    try:
        prediction = predictor.predict(input_data.dict())
        return {"congestion_level": prediction}
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

Step 5: Create AWS Infrastructure

Create "infrastructure.py":

import boto3
import json

def create_infrastructure():
    # Create ECR repository
    ecr = boto3.client('ecr')
    try:
        ecr.create_repository(repositoryName='traffic-predictor')
    except ecr.exceptions.RepositoryAlreadyExistsException:
        pass

    # Create ECS cluster
    ecs = boto3.client('ecs')
    ecs.create_cluster(clusterName='traffic-predictor-cluster')

    # Create task definition
    task_def = {
        'family': 'traffic-predictor',
        'containerDefinitions': [{
            'name': 'traffic-predictor',
            'image': f'{ecr.describe_repositories()["repositories"][0]["repositoryUri"]}:latest',
            'memory': 512,
            'cpu': 256,
            'essential': True,
            'portMappings': [{
                'containerPort': 8000,
                'hostPort': 8000,
                'protocol': 'tcp'
            }]
        }],
        'requiresCompatibilities': ['FARGATE'],
        'networkMode': 'awsvpc',
        'cpu': '256',
        'memory': '512'
    }

    ecs.register_task_definition(**task_def)

Step 6: Containerise the Application

Create "Dockerfile:"

FROM python:3.9-slim

WORKDIR /app

COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

COPY . .

CMD ["uvicorn", "api:app", "--host", "0.0.0.0", "--port", "8000"]

Create "requirements.txt:"

fastapi
uvicorn
boto3
pandas
numpy
scikit-learn

Step 7: Deploy to AWS

Run these commands:

# Build and push Docker image
aws ecr get-login-password --region us-east-1 | docker login --username AWS --password-stdin $AWS_ACCOUNT_ID.dkr.ecr.us-east-1.amazonaws.com
docker build -t traffic-predictor .
docker tag traffic-predictor:latest $AWS_ACCOUNT_ID.dkr.ecr.us-east-1.amazonaws.com/traffic-predictor:latest
docker push $AWS_ACCOUNT_ID.dkr.ecr.us-east-1.amazonaws.com/traffic-predictor:latest

# Create infrastructure
python infrastructure.py

Step 8: Update Streamlit Frontend

Modify "app.py" to connect to the API:

import streamlit as st
import requests
import plotly.graph_objects as go
import plotly.express as px

API_ENDPOINT = "your-api-endpoint"

def predict_traffic(input_data):
    response = requests.post(f"{API_ENDPOINT}/predict", json=input_data)
    return response.json()["congestion_level"]

# Rest of the Streamlit code remains the same, but replace direct model calls
# with API calls using predict_traffic()

Step 9: Testing and Monitoring

Test the API endpoint:

curl -X POST "your-api-endpoint/predict" \
     -H "Content-Type: application/json" \
     -d '{"hour":12,"day":1,"temperature":25,"precipitation":0,"special_event":false,"road_work":false,"vehicle_count":1000}'

Monitor using AWS CloudWatch:

• Set up CloudWatch dashboard
• Create alarms for error rates and latency
• Monitor API usage and costs

If everything goes well. Congratulations! You've successfully deployed a traffic congestion predictor. Pad yourself on the back for that one! Make sure you monitor costs and performance, regularly update the model, and implement a CI/CD pipeline. The next steps are adding user authentication, enhancing monitoring and alerting, optimising model performance, and adding more features based on user feedback.

Thanks for reading this. Let me know any thoughts, questions or observations!