Skip to main content

█▄ █ █▀▀ █ █ █ ▀█▀ █ █
█ ▀█ █▀▀ ▀▄▀▄▀  █  █▀█
█  █ █▄▄  ▀ ▀   █  █ █
  ▄▀█ █
  █▀█ █
°

Loading...

Preparing your experience

© 2025 Oliver Newth. All rights reserved.

Projects Work GitHub LinkedIn

Case Studies

Outcomes-first narratives showing how I've built and scaled AI systems at Google, Meta, Microsoft, and startups. Real numbers, real decisions, real tradeoffs.

3 case studies

Billions of users impacted

42 minutes of reading

Featured

Instagram Calling: 0 to 75% DAU in 6 Months

How we launched real-time calling to billions of users without breaking Instagram

Launched Instagram Calling from zero to 75% daily active user adoption in 6 months while maintaining platform stability and trust & safety standards.

Call Completion Rate

vs 70% industry avg

Harmful Content

0.16% → 0.12%

15 min readRead case study →

Covariant (acquired by Amazon)

Warehouse Robotics: 72% to 85% Accuracy, -20% Safety Incidents

How we made warehouse robots reliable enough for Amazon to acquire us

Improved object detection accuracy from 72% to 85% in cluttered warehouse environments while reducing safety incidents by 20% across deployments.

Object Detection Accuracy

Safety Incidents

baseline → -20%

across deployments

Edge Case Accuracy

baseline → +8%

objects per hour

14 min readRead case study →

ML Platform: 6 Hours to 80 Minutes Deployment, 99.99% Uptime

How we made ML deployment fast and reliable for 30+ product teams

Reduced model deployment time from 6 hours to 80 minutes across the organization while maintaining 99.99% platform uptime over 2 years.

Deployment Time

6 hours → 80 minutes

Platform Uptime

99.5% → 99.99%

Deployments per Week

Mean Time to Recovery

30 min → 5 min

13 min readRead case study →

Product

Meta•2017-2022

Instagram Calling: 0 to 75% DAU in 6 Months

How we launched real-time calling to billions of users without breaking Instagram

DAU Adoption:0% → 75%

Daily Calls:0 → 600M

Call Completion Rate:N/A → 85%

Infrastructure

Covariant (acquired by Amazon)•2022-2024

Warehouse Robotics: 72% to 85% Accuracy, -20% Safety Incidents

How we made warehouse robots reliable enough for Amazon to acquire us

Object Detection Accuracy:72% → 85%

Safety Incidents:baseline → -20%

Edge Case Accuracy:45% → 78%

Google•2025-Present

ML Platform: 6 Hours to 80 Minutes Deployment, 99.99% Uptime

How we made ML deployment fast and reliable for 30+ product teams

Deployment Time:6 hours → 80 minutes

Platform Uptime:99.5% → 99.99%

Deployments per Week:50 → 200

Want similar results for your team?

I work with teams to implement these patterns and achieve measurable outcomes.

Advisory Services See Frameworks