BytesByBytes - Building technology, one byte at a time.

BytesByBytes - Building technology, one byte at a time.Practical notes, experiments and deep dives into software, AI, data and embedded systems.https://bytesbybytes.pages.deven-usWhy BLE Disconnects When Your Android Screen Turns Offhttps://bytesbybytes.pages.dev/articles/android-ble-background-disconnecthttps://bytesbybytes.pages.dev/articles/android-ble-background-disconnectA practical investigation into Android background restrictions, BLE connection behaviour and how to make connections survive a locked screen.Sat, 29 Aug 2026 00:00:00 GMTMobile DevelopmentBLEAndroidReact NativeGourav PradhanBuilding AI at the Edge: What Developers Actually Need to Knowhttps://bytesbybytes.pages.dev/articles/building-ai-at-the-edgehttps://bytesbybytes.pages.dev/articles/building-ai-at-the-edgeRunning models on devices with no cloud to fall back on changes almost every assumption you carry over from server-side inference. Here is what actually matters.Tue, 17 Feb 2026 00:00:00 GMTEdge AIEdge AIQuantizationEmbeddedInferenceGourav PradhanStructured Outputs Are Not an Evaluation Strategyhttps://bytesbybytes.pages.dev/articles/structured-outputs-are-not-evaluationhttps://bytesbybytes.pages.dev/articles/structured-outputs-are-not-evaluationConstrained decoding guarantees your model returns valid JSON. It guarantees nothing about whether the values inside it are right - and the two get confused constantly.Tue, 30 Sep 2025 00:00:00 GMTAI & Machine LearningLLMEvaluationJSONGourav PradhanWhere React Native Cold Start Time Actually Goeshttps://bytesbybytes.pages.dev/articles/react-native-cold-starthttps://bytesbybytes.pages.dev/articles/react-native-cold-startCold start is rarely one slow thing. It is a dozen small ones, and profiling it properly changes which of them you bother to fix.Wed, 21 May 2025 00:00:00 GMTMobile DevelopmentReact NativeAndroidPerformanceGourav PradhanCI Caching That Actually Workshttps://bytesbybytes.pages.dev/articles/ci-caching-that-actually-workshttps://bytesbybytes.pages.dev/articles/ci-caching-that-actually-worksMost CI caches are slower than no cache at all. The difference comes down to what you key on, what you store and how honestly you measure the result.Tue, 19 Nov 2024 00:00:00 GMTDeveloper ToolsCIBuild SystemsPerformanceGourav PradhanIncremental Pipelines That Can Be Run Twicehttps://bytesbybytes.pages.dev/articles/idempotent-incremental-pipelineshttps://bytesbybytes.pages.dev/articles/idempotent-incremental-pipelinesIdempotency is the property that turns a fragile nightly job into something you can retry without thinking. Here is how to build it into a pipeline from the start.Thu, 05 Sep 2024 00:00:00 GMTData EngineeringPipelinesSQLReliabilityGourav PradhanHow Parquet File Layout Decides Your Query Speedhttps://bytesbybytes.pages.dev/articles/parquet-layout-query-speedhttps://bytesbybytes.pages.dev/articles/parquet-layout-query-speedTwo datasets with identical rows and identical schemas can differ by an order of magnitude in scan time. The difference is layout - row groups, sort order and file size.Thu, 27 Jun 2024 00:00:00 GMTData EngineeringParquetStoragePerformanceGourav PradhanError Handling That Survives Contact With Productionhttps://bytesbybytes.pages.dev/articles/errors-that-survive-productionhttps://bytesbybytes.pages.dev/articles/errors-that-survive-productionMost error handling is written for the developer who wrote it, on the day they wrote it. Here is what makes it useful six months later at 3 a.m.Tue, 12 Mar 2024 00:00:00 GMTSoftware DevelopmentReliabilityLoggingDesignGourav Pradhan