AWS re:Invent 2025 - High-performance NLP & geospatial analysis with Amazon Redshift (ANT334)

Summary of AWS re:Invent 2025 - High-performance NLP & Geospatial Analysis with Amazon Redshift (ANT334)

Overview

• Presentation by Ryan McMahon from Cambridge Mobile Telematics (CMT), a technology company that measures driving behavior to improve road safety
• Discusses how CMT is leveraging AWS services, particularly Amazon Redshift, to expand their capabilities from individual driver risk analysis to broader geospatial risk identification

Evolving Challenges in Road Safety

• CMT's initial focus was on understanding individual driver behavior and risks to prevent crashes and fatalities
• Over time, the data volume and complexity has grown significantly as they incorporate more sensor data from smartphones, proprietary tags, and other platforms
• Realized that to truly reduce road fatalities, they need to look beyond just individual driver risk and also consider vehicle and road-related factors

Geospatial Risk Analysis with Amazon Redshift

• Shifted focus to identifying and predicting where crashes are most likely to occur, rather than just reacting to individual incidents
• Needed a scalable platform to process the massive volume of data (over 1 trillion data points per day) and perform advanced geospatial analysis
• Chose to leverage Amazon Redshift for its:
  • Massively Parallel Processing (MPP) architecture
  • Ability to handle petabyte-scale data
  • Native support for spatial data types and over 100 spatial functions

Leveraging H3 Hexagonal Indexing

• Discovered that traditional spatial queries using polygons and bounding boxes were inefficient at scale
• Adopted the H3 hexagonal hierarchical indexing system, which provides:
  • Pre-computed spatial indexes for fast lookups
  • Ability to choose the desired resolution level (16 levels from 0 to 15)
  • Up to 98% reduction in compute resources compared to traditional spatial queries

Integrating NLP Capabilities

• Utilized Amazon Redshift's integration with Amazon QSQL, which provides natural language processing (NLP) capabilities
• Allows non-SQL experts to use plain English to query the data warehouse
• Leverages the full context of the Redshift schema and query history to generate accurate SQL queries

Business Impact and Use Cases

• Identified high-risk road segments, such as a location in Arkansas where a stop sign was obstructed by vegetation
• Able to proactively address these issues before crashes occur, rather than waiting for incidents to happen
• Described the goal of creating a "Zillow for road safety" - a platform that can identify and visualize road safety risks globally

Key Takeaways

• CMT's partnership with AWS, particularly Amazon Redshift, has enabled them to scale their geospatial risk analysis capabilities
• The integration of H3 hexagonal indexing and NLP-powered querying has significantly improved the performance and accessibility of their platform
• By shifting from reactive to proactive road safety measures, CMT aims to make a meaningful impact in reducing road fatalities worldwide