Notes on the architecture of artificial intelligence systems: design, engineering, operations, and the practice of building reliable solutions.https://maxbirkin.com/enhttps://maxbirkin.com/blog/latency-engineering/https://maxbirkin.com/blog/latency-engineering/How a set of deliberate levers, from parallelism to merging two model calls into one response, cut the full log-analysis time by an order of tens.Fri, 10 Jul 2026 00:00:00 GMThttps://maxbirkin.com/blog/setfit-query-routing/https://maxbirkin.com/blog/setfit-query-routing/Why a lightweight trained classifier stands between the code and the large model, how SetFit learns from a handful of examples, and when a query escalates to the LLM.Thu, 09 Jul 2026 00:00:00 GMThttps://maxbirkin.com/blog/cascade-chains/https://maxbirkin.com/blog/cascade-chains/Why five affected services are most often one root cause plus four relays of its failure, and how cascade analysis narrows the hypothesis space.Tue, 07 Jul 2026 00:00:00 GMThttps://maxbirkin.com/blog/no-quote-no-claim/https://maxbirkin.com/blog/no-quote-no-claim/Anti-hallucination guards for open-ended root cause analysis: multiplying the model's self-assessment by the share of its evidence actually found in the sources.Sat, 04 Jul 2026 00:00:00 GMThttps://maxbirkin.com/blog/bayesian-confidence-scoring/https://maxbirkin.com/blog/bayesian-confidence-scoring/How to build verdict confidence from triggered detectors with deterministic code, without trusting a number from the language model, and why group-max beats the product.Wed, 01 Jul 2026 00:00:00 GMThttps://maxbirkin.com/blog/single-pass-deterministic-aggregator/https://maxbirkin.com/blog/single-pass-deterministic-aggregator/How to remove the language model from the log-chunking stage and replace it with a statistical aggregator that computes 16 blocks of facts in one linear pass.Wed, 24 Jun 2026 00:00:00 GMThttps://maxbirkin.com/blog/inversion-of-responsibility/https://maxbirkin.com/blog/inversion-of-responsibility/The central principle of a mature log-analysis system: divide responsibility between deterministic code, a trained classifier, and a large language model.Tue, 16 Jun 2026 00:00:00 GMT