October 17, 2026
Dear [Hiring manager's name],
At Parcelwise I own the lifecycle of seven production models, courier dispatch, ETA prediction and fraud screening, that run underneath about 2 million orders a month, and the part of the job I care most about is what happens after a model ships, not the demo. I'm drawn to your Machine Learning Engineer opening because it treats production ownership, monitoring, retraining, rollback, as core to the role instead of someone else's problem downstream.
Compressing our ETA transformer through distillation and dynamic batching on Triton took its p99 inference latency from 210 milliseconds to 45 and cut GPU serving cost roughly in half, and reframing our dispatch ranking objective lifted courier acceptance rate 12%, validated through a six-arm A/B test over three weeks rather than a single before-and-after snapshot.
I built drift detection across 300 features that caught a payments-schema change before it could quietly cost our fraud model an estimated 9 points of recall, and I run weekly retraining on Kubeflow behind shadow-evaluation gates so a challenger model only promotes after beating the champion on held-out data. I work daily in PyTorch and TensorFlow, hold the AWS Certified Machine Learning - Specialty and Google Cloud Professional Machine Learning Engineer credentials, and I'd rather add one more monitoring check than ship a model I can't explain when it drifts.
I'd welcome a conversation about the models [Company name] is running in production today and where six years of shipping and maintaining them, not just training them, could help most. Thank you for your consideration.
Sincerely,
Peyton Clark
