diff --git a/.gitignore b/.gitignore
index 27a7411..0759349 100644
--- a/.gitignore
+++ b/.gitignore
@@ -47,6 +47,7 @@ serverless/*/src/*
 !serverless/*/src/alpr_counts.py
 !serverless/*/src/alpr_clusters.py
 !serverless/*/src/requirements.txt
+!serverless/*/src/Dockerfile
 
 # TODO: need a better way to handle python packages
 
diff --git a/serverless/alpr_clusters/src/Dockerfile b/serverless/alpr_clusters/src/Dockerfile
new file mode 100644
index 0000000..d80bd11
--- /dev/null
+++ b/serverless/alpr_clusters/src/Dockerfile
@@ -0,0 +1,12 @@
+# Use the official AWS Lambda Python 3.9 base image
+FROM amazon/aws-lambda-python:3.9
+
+# Copy function code
+COPY alpr_clusters.py ${LAMBDA_TASK_ROOT}
+
+# Install dependencies
+COPY requirements.txt .
+RUN pip install -r requirements.txt
+
+# Set the CMD to your handler
+CMD ["alpr_clusters.lambda_handler"]
diff --git a/serverless/alpr_clusters/src/alpr_clusters.py b/serverless/alpr_clusters/src/alpr_clusters.py
index f00b092..56bee4f 100644
--- a/serverless/alpr_clusters/src/alpr_clusters.py
+++ b/serverless/alpr_clusters/src/alpr_clusters.py
@@ -1,10 +1,45 @@
-import boto3
-import requests
 import json
-from sklearn.cluster import DBSCAN
-import numpy as np
+from collections import defaultdict
+from typing import Any
 
-def get_clusters():
+import boto3
+import os
+import time
+import numpy as np
+import requests
+from sklearn.cluster import DBSCAN
+from typing import Tuple
+import re
+from concurrent.futures import ThreadPoolExecutor
+
+def terraform_rate_expression_to_minutes(rate_expression: str) -> Tuple[int, int]:
+    match = re.match(r"rate\((\d+)\s*(day|hour|minute)s?\)", rate_expression)
+    if not match:
+        raise ValueError(f"Invalid rate expression: {rate_expression}")
+
+    value, unit = int(match.group(1)), match.group(2)
+
+    if unit == "day":
+        return value * 24 * 60
+    elif unit == "hour":
+        return value * 60
+    elif unit == "minute":
+        return value
+    else:
+        raise ValueError(f"Unsupported time unit: {unit}")
+    
+UPDATE_RATE_MINS = terraform_rate_expression_to_minutes(os.getenv("UPDATE_RATE_MINS", "rate(60 minutes)"))
+GRACE_PERIOD_MINS = int(2) # XXX: set expiration a few minutes after in case put object takes a while
+TILE_SIZE_DEGREES = int(20)
+
+WHITELISTED_TAGS = [
+    "operator",
+    "manufacturer",
+    "direction",
+    "brand",
+]
+
+def get_all_nodes():
     # Set up the Overpass API query
     query = """
     [out:json];
@@ -12,31 +47,40 @@ def get_clusters():
     out body;
     """
 
+    url = "http://overpass-api.de/api/interpreter"
+    response = requests.get(
+        url, params={"data": query}, headers={"User-Agent": "DeFlock/1.0"}
+    )
+    response.raise_for_status()
+    return response.json()["elements"]
+
+
+def get_clusters(nodes: list[Any]):
     # Request data from Overpass API
     print("Requesting data from Overpass API.")
-    url = "http://overpass-api.de/api/interpreter"
-    response = requests.get(url, params={'data': query}, headers={'User-Agent': 'DeFlock/1.0'})
-    data = response.json()
+
     print("Data received. Parsing nodes...")
 
     # Parse nodes and extract lat/lon for clustering
     coordinates = []
     node_ids = []
-    for element in data['elements']:
-        if element['type'] == 'node':
-            coordinates.append([element['lat'], element['lon']])
-            node_ids.append(element['id'])
+    for element in nodes:
+        if element["type"] == "node":
+            coordinates.append([element["lat"], element["lon"]])
+            node_ids.append(element["id"])
 
     # Convert coordinates to NumPy array for DBSCAN
     coordinates = np.array(coordinates)
 
-    # Define the clustering radius (10 miles in meters)
+    # Define the clustering radius (50 miles in meters)
     radius_miles = 50
     radius_km = radius_miles * 1.60934  # 1 mile = 1.60934 km
     radius_in_radians = radius_km / 6371.0  # Earth's radius in km
 
     # Perform DBSCAN clustering
-    db = DBSCAN(eps=radius_in_radians, min_samples=1, algorithm='ball_tree', metric='haversine').fit(np.radians(coordinates))
+    db = DBSCAN(
+        eps=radius_in_radians, min_samples=1, algorithm="ball_tree", metric="haversine"
+    ).fit(np.radians(coordinates))
     labels = db.labels_
 
     # Prepare clusters and calculate centroids
@@ -45,34 +89,108 @@ def get_clusters():
         cluster_points = coordinates[labels == label]
         centroid = np.mean(cluster_points, axis=0)
         first_node_id = node_ids[labels.tolist().index(label)]
-        
+
         # Store in clusters dict with centroid and first node ID
-        clusters[label] = {
-            "lat": centroid[0],
-            "lon": centroid[1],
-            "id": first_node_id
-        }
+        clusters[label] = {"lat": centroid[0], "lon": centroid[1], "id": first_node_id}
 
     output = {"clusters": list(clusters.values())}
     print("Clustering complete.")
     return output
 
 
+def segment_regions(nodes: Any, tile_size_degrees: int) -> dict[Any]:
+    print("Segmenting regions...")
+    tile_dict = defaultdict(list)
+    for node in nodes:
+        lat, lon = node["lat"], node["lon"]
+        tile_lat = int(np.floor(lat / tile_size_degrees)) * tile_size_degrees
+        tile_lon = int(np.floor(lon / tile_size_degrees)) * tile_size_degrees
+        bare_node = {
+            "id": node["id"],
+            "lat": lat,
+            "lon": lon,
+            "tags": {k: v for k, v in node["tags"].items() if k in WHITELISTED_TAGS},
+        }
+        tile_dict[f"{tile_lat}/{tile_lon}"].append(bare_node)
+    print("Region segmentation complete.")
+
+    return tile_dict
+
 def lambda_handler(event, context):
-  alpr_clusters = get_clusters()
+    nodes = get_all_nodes()
+    alpr_clusters = get_clusters(nodes)
+    regions_dict = segment_regions(nodes=nodes, tile_size_degrees=TILE_SIZE_DEGREES)
 
-  s3 = boto3.client('s3')
-  bucket = 'deflock-clusters'
-  key = 'alpr_clusters.json'
+    print("Uploading data to S3...")
 
-  s3.put_object(
-    Bucket=bucket,
-    Key=key,
-    Body=json.dumps(alpr_clusters),
-    ContentType='application/json'
-  )
+    s3 = boto3.client("s3")
+    bucket = "deflock-clusters"
+    bucket_new = "cdn.deflock.me"
+    key = "alpr_clusters.json"
 
-  return {
-    'statusCode': 200,
-    'body': 'Successfully fetched ALPR counts.',
-  }
+    s3.put_object(
+        Bucket=bucket,
+        Key=key,
+        Body=json.dumps(alpr_clusters),
+        ContentType="application/json",
+    )
+
+    # TODO: handle outdated index files when their referenced files are deleted
+    epoch = int(time.time())
+    tile_index = {
+        "expiration_utc": epoch + (UPDATE_RATE_MINS + GRACE_PERIOD_MINS) * 60,
+        "regions": list(regions_dict.keys()),
+        "tile_url": "https://cdn.deflock.me/regions/{lat}/{lon}.json?v=" + str(epoch),
+        "tile_size_degrees": TILE_SIZE_DEGREES,
+    }
+
+    print("Uploading regions to S3...")
+
+    def upload_json_to_s3(bucket, key, body):
+        s3.put_object(
+            Bucket=bucket,
+            Key=key,
+            Body=body,
+            ContentType="application/json",
+        )
+
+    # Use ThreadPoolExecutor for concurrent uploads
+    with ThreadPoolExecutor(max_workers=10) as executor:
+        futures = []
+        for latlng_string, elements in regions_dict.items():
+            lat, lon = latlng_string.split("/")
+            key = f"regions/{lat}/{lon}.json"
+            body = json.dumps(elements)
+            futures.append(executor.submit(upload_json_to_s3, bucket_new, key, body))
+        
+        # add index file
+        futures.append(executor.submit(upload_json_to_s3, bucket_new, "regions/index.json", json.dumps(tile_index)))
+
+        # Wait for all futures to complete
+        for future in futures:
+            future.result()
+
+    print("Regions uploaded to S3. Done!")
+
+    return {
+        "statusCode": 200,
+        "body": "Successfully clustered.",
+    }
+
+
+if __name__ == "__main__":
+    from pathlib import Path
+
+    nodes_file_path = Path("nodes.json")
+    if nodes_file_path.exists():
+        nodes = json.load(nodes_file_path.open())
+    else:
+        nodes = get_all_nodes()
+        with nodes_file_path.open("w") as f:
+            json.dump(nodes, f)
+
+
+
+    regions_dict = segment_regions(nodes=nodes, tile_size_degrees=5)
+    with open("regions_dict.json", "w") as f:
+        json.dump(regions_dict, f)
diff --git a/terraform/main.tf b/terraform/main.tf
index e9b2896..b219fe1 100644
--- a/terraform/main.tf
+++ b/terraform/main.tf
@@ -13,5 +13,6 @@ module "alpr_clusters" {
   module_name = "alpr_clusters"
   source = "./modules/alpr_clusters"
   deflock_stats_bucket = var.deflock_stats_bucket
+  deflock_cdn_bucket = var.deflock_cdn_bucket
   rate = "rate(1 day)"
 }
diff --git a/terraform/modules/alpr_clusters/main.tf b/terraform/modules/alpr_clusters/main.tf
index 2488460..fc3e989 100644
--- a/terraform/modules/alpr_clusters/main.tf
+++ b/terraform/modules/alpr_clusters/main.tf
@@ -15,6 +15,11 @@ resource "aws_iam_role" "lambda_role" {
   })
 }
 
+resource "aws_iam_role_policy_attachment" "lambda_basic_execution" {
+  role       = aws_iam_role.lambda_role.name
+  policy_arn = "arn:aws:iam::aws:policy/service-role/AWSLambdaBasicExecutionRole"
+}
+
 resource "aws_iam_policy" "lambda_s3_write_policy" {
   name        = "${var.module_name}_lambda_s3_write_policy"
   description = "Policy for Lambda to write to S3 bucket ${var.deflock_stats_bucket}"
@@ -28,7 +33,10 @@ resource "aws_iam_policy" "lambda_s3_write_policy" {
           "s3:PutObjectAcl"
         ]
         Effect   = "Allow"
-        Resource = "arn:aws:s3:::${var.deflock_stats_bucket}/${var.output_filename}"
+        Resource = [
+          "arn:aws:s3:::${var.deflock_cdn_bucket}/*",
+          "arn:aws:s3:::${var.deflock_stats_bucket}/*"
+        ] 
       }
     ]
   })
@@ -44,7 +52,13 @@ resource "aws_lambda_function" "overpass_lambda" {
   role             = aws_iam_role.lambda_role.arn
   package_type     = "Image"
   image_uri        = "${aws_ecr_repository.lambda_repository.repository_url}:latest"
-  timeout = 90
+  timeout = 180
+  memory_size = 512
+  environment {
+    variables = {
+      UPDATE_RATE_MINS = var.rate
+    }
+  }
 }
 
 resource "aws_cloudwatch_event_rule" "lambda_rule" {
@@ -70,3 +84,8 @@ resource "aws_lambda_permission" "allow_cloudwatch_to_call_lambda" {
 resource "aws_ecr_repository" "lambda_repository" {
   name = "${var.module_name}-lambda"
 }
+
+resource "aws_cloudwatch_log_group" "lambda_log_group" {
+  name              = "/aws/lambda/${aws_lambda_function.overpass_lambda.function_name}"
+  retention_in_days = 14
+}
diff --git a/terraform/modules/alpr_clusters/variables.tf b/terraform/modules/alpr_clusters/variables.tf
index 98d92fa..ab6b9b1 100644
--- a/terraform/modules/alpr_clusters/variables.tf
+++ b/terraform/modules/alpr_clusters/variables.tf
@@ -11,6 +11,10 @@ variable "deflock_stats_bucket" {
   description = "S3 bucket for the ALPR clusters JSON file"
 }
 
+variable "deflock_cdn_bucket" {
+  description = "S3 bucket for the CDN"
+}
+
 variable "rate" {
   description = "Rate at which to run the Lambda function"
 }
diff --git a/terraform/variables.tf b/terraform/variables.tf
index b0f2d59..28ab153 100644
--- a/terraform/variables.tf
+++ b/terraform/variables.tf
@@ -2,3 +2,8 @@ variable "deflock_stats_bucket" {
   description = "S3 bucket for the ALPR counts JSON file"
   default = "deflock-clusters"
 }
+
+variable "deflock_cdn_bucket" {
+  description = "S3 bucket for the CDN"
+  default = "cdn.deflock.me"
+}