Works on retrieval infrastructure. Writes about the parts of ML that turn out to be distributed systems problems.https://strata.vedant31.techenhttps://strata.vedant31.tech/w/why-your-embeddings-are-worse-than-you-thinkhttps://strata.vedant31.tech/w/why-your-embeddings-are-worse-than-you-thinkDimension buys separation, not meaning. The curve flattens hard, and three pipeline bugs were costing more than the model.Wed, 24 Jun 2026 16:59:29 GMT2Machine LearningVedanthttps://strata.vedant31.tech/w/we-deleted-our-staging-environmenthttps://strata.vedant31.tech/w/we-deleted-our-staging-environmentSix months in. The best thing we did was delete a database nobody had ever tested against.Wed, 01 Apr 2026 16:59:29 GMT1SystemsVedanthttps://strata.vedant31.tech/w/the-p99-is-a-lie-you-tell-yourselfhttps://strata.vedant31.tech/w/the-p99-is-a-lie-you-tell-yourselfYou cannot add percentiles. Every latency budget that does is a budget nobody is actually keeping.Mon, 02 Mar 2026 16:59:29 GMT2SystemsVedanthttps://strata.vedant31.tech/w/field-guide-to-evaluating-retrievalhttps://strata.vedant31.tech/w/field-guide-to-evaluating-retrievalRecall@k tells you if you found the chunk. It tells you nothing about whether the chunk was useful.Fri, 27 Dec 2024 16:59:29 GMT3Machine LearningVedanthttps://strata.vedant31.tech/w/cache-invalidation-is-a-distributed-problemhttps://strata.vedant31.tech/w/cache-invalidation-is-a-distributed-problemThe TTL is not a cache policy. It is a probability distribution, and somebody has to own it.Mon, 13 Nov 2023 16:59:29 GMT4SystemsVedant