Timely analysis, tutorials, and opinion from ScaleCloud architects on the cloud and AI landscape.
10 resources
Pure vector similarity search retrieves semantically similar chunks, but similarity isn't the same as relevance — a chunk can be topically similar to a…
Use structure-aware chunking (respecting headings, tables, and document sections) rather than fixed-length character chunking, which materially improves retrieval quality on struct
Rightsizing requests and limits based on actual usage (not guesses) is table stakes, but it typically saves 10-20% — the bigger wins come from…
Use Azure Network Watcher's connection troubleshoot tool alongside your connectivity provider's own monitoring; also confirm you haven't inadvertently created asymmetric routing be
Object storage (S3/ADLS) holds the physical data, with Apache Iceberg as the open table format layer providing ACID transactions, schema evolution, and time travel…
Avoid granting Owner or subscription-level Contributor purely for convenience — scope the service principal to the specific resource group(s) it manages, and rotate its…
Centralizing egress, firewall inspection, and shared services (DNS, identity, monitoring) in the hub avoids duplicating security infrastructure per workload team while still giving
Before extracting any service, map your monolith's internal module boundaries and data ownership — the seams that already exist in your codebase (even informally)…
