Faster predictions with C++

I transpiled the Python-based prediction scripts into C++ and packaged them as distributable binaries. For property predictions made on on-premises infrastructure, this gave a speedup of 1.6X.

A pull-based prediction engine

I refactored the prediction engine so that workers pull jobs from SQS queues instead of having work pushed to them. This improved hourly job throughput in production by 43%. It also let us remove unmaintained dependencies from production builds.

Refactored prediction engine to consume jobs from SQS queues, improving job throughput in production by 43% and eliminating unmaintained dependencies from production builds.

Automated backup testing and restoration

I automated production backup testing and restoration with Terraform. Testing coverage went from less than 1% of platform data to 100%, and recovery takes under 20 minutes.

Automated production backup testing and restoration using Terraform, increasing testing coverage from less than 1% of platform data to 100% and achieving a recovery time of under 20 minutes.