CandidPartners → McKinsey (2020–2021)
I joined CandidPartners in May 2020, embedded in an existing DevOps team at Repay.com, a large credit card processing company. On January 1, 2021, McKinsey acquired CandidPartners — my role continued without interruption.
Repay.com — Payment Processing Platform
Stack: C# on .NET (not .NET Core) → Windows EC2 via Auto Scaling Groups, MSSQL on RDS (multi-region), CloudFront for static assets.
CloudFront Geo-Based Load Balancing
In July 2020, CloudFront released geolocation headers — GPS coordinates injected into each request. I modified the existing Edge Lambdas to use the client's longitude to deterministically route requests to either us-east-1 or us-west-2:
region = 'us-east-1'
if (longitude < us_meridian) {
region = 'us-west-2'
}
return region
The threshold was stored in an SSM Parameter (US-Meridian) for easy tuning. The result: more deterministic load distribution and reduced pressure on the inter-region replication layer during lag events.
Batch CC Transaction Encryption via Lambda
Built a highly parallelized Lambda function to encrypt recently-cleared credit card transactions as a background task. Rather than one-record-at-a-time (the typical Lambda tutorial model), I used the full 15-minute execution window — encrypting thousands of transactions per invocation, monitoring elapsed time, and committing/closing cleanly at the 14.5-minute mark. Batch processing inside Lambda, essentially.
Federal Reserve Bank — ELMA
During the CandidPartners engagement, I also served the Federal Reserve Bank, where we built ELMA (Enterprise Logging Metrics and Analysis) — a serverless log ingestion product to normalize ~20 different log formats:
- Preprocessing Lambdas ingested and normalized raw logs
- Amazon Kinesis Stream + Kinesis Firehose wrote data to an AWS Glue database (S3-backed)
- AWS QuickSight enabled all users to query all log types as a unified SQL table