A LITTLE ROOM TO THINK
Worth a thought.
Notes on life, design, and the digital world. Written with care, for the curious.
Latest stories
Time-Series Validation: Preventing Leakage from Late Data and Future Labels
Build time-series validation around feature availability and label cutoffs. A reproducible Python experiment exposes leakage that chronological splits cannot catch.
Redis Cache Stampedes: Single Flight, TTL Jitter, and Stale-While-Revalidate
Control Redis cache stampedes with a tested Python refresh lease, TTL jitter, and bounded stale-data policies. Includes a reproducible lab and 15 integration tests.
Kafka Offset Commits with Parallel Workers: Track the Completed Prefix
A tested Python model shows how to commit Kafka offsets safely when workers finish out of order, with offset gaps, bounded queues, and assignment changes.
Evaluating RAG Retrieval Before Changing the Language Model
Measure RAG evidence with Recall@k, MRR, and nDCG, using tested Python code to expose missing passages, ranking tradeoffs, and context-packing failures.