Writing
Your Code Review Agent Needs Eval
Generic benchmarks can't tell you whether a review agent works on your codebase. Here's how to build an eval from your own PR history, roll it up to a single ship-or-block decision, and run it sealed with Harbor.
Classify Jira Tickets with GenAI on Amazon Bedrock
Replace traditional NLP approaches with prompt engineering and large language models for Jira ticket text classification. A code sample walkthrough.
Evaluating LLM Systems: Best Practices
A guide to implementing evaluation frameworks across every stage of LLM-augmented system development, from embeddings through production monitoring.
Optimizing CTR Model Training on 1B+ Datapoints
How I cut a click-through rate model's training time from ~2 days to 15 hours on 3TB of tabular data in TensorFlow 2, by working through four bottlenecks — I/O, data loader preprocessing, GPU communication overhead, and kernel launch times.
Run a PyTorch Model on AWS Inferentia2
A hands-on walkthrough of training a small multilayer perceptron (MLP) to predict California housing prices, then compiling it with the AWS Neuron SDK and running inference on an Amazon EC2 inf2.xlarge instance.