Posts

Implement atomic tool-execution claims in .NET with EF Core, separate dispatch intents from attempts, and recover lost responses through stored audit evidence.

Enforce AI telemetry privacy with versioned capture grants, atomic quotas across replicas, and evidence expiry that survives queues and evaluation copies.

Compare cache, retry, and telemetry order around IChatClient with a fake provider, exact event traces, cancellation, and a streaming failure.

Evaluate real Qdrant results in .NET with a custom IEvaluator, labeled chunks, Precision@K, Recall@K, MRR, nDCG, and persisted HTML reports.

Assemble model context from Qdrant results in .NET: deduplicate exact text, check a local token allowance, and trace citations to admitted evidence.

Extend the .NET Qdrant chunking sample with payload metadata, keyword indexes, and filters that restrict retrieval to published runbooks.

Use admission control, bounded waiting, and shared provider limits to keep slow AI work from turning overload into an outage, with C# examples.

Extend .NET and Qdrant indexing with section boundaries, paragraph overlap, deterministic chunk IDs, and queries that expose incomplete evidence.

Qualify fallback models against the feature contract, define eligible failures, bound attempts, and make reduced behavior visible to callers.

Compose shared chat-client behavior in .NET without moving feature policy into middleware or hiding streaming, caching, and retry costs.