skip to content

Search

Posts RSS feed

2025

  • Build a streaming anomaly detection pipeline with Kafka and PyTorch. Train an autoencoder on normal sensor behavior, then deploy it for real-time inference on live data.

  • Move beyond vibes-based testing. Build a systematic evaluation harness for LLM applications using deterministic checks, semantic similarity, and LLM-as-judge patterns.

2024

  • Why sampling is a modeling problem, not a convenience, and how to construct datasets that are both workable and representative using stratified sampling strategies

  • A practical, production-ready walkthrough for tuning XGBoost with Optuna: strong search spaces, pruning, reproducibility, and holdout-safe evaluation.