Blog
Artwork: Irises by Vincent van Gogh. The Metropolitan Museum of Art · Public domain
AI AgentsHybrid Activity Classification: Keyword Rules First, LLM Fallback
Keyword rules handle clear classifications, while an LLM receives only ambiguous cases. A confidence threshold controls escalation, reducing cost and latency without surrendering deterministic decisions.
Artwork: Aristotle with a Bust of Homer by Rembrandt. The Metropolitan Museum of Art · Public domain
AI AgentsReal-Time Voice Debriefing with the Azure Voice Live API
A persistent WebSocket supports an eight-step hands-free debrief, while persona, scenario, and environment fixtures make the nondeterministic voice agent repeatable enough to evaluate.
Artwork: Diana and Her Nymphs Departing for the Hunt by Peter Paul Rubens. Cleveland Museum of Art · Public domain
AI AgentsDesigning a Multi-Agent Field Service Assistant
Specialized briefing, debriefing, and classification agents behind an orchestrator are easier to prompt, test, and evaluate than one large agent, while connectors keep local and cloud modes replaceable.
Artwork: Marcelle la Blonde by Juan Gris. Cleveland Museum of Art · Public domain
RAGFour Chunking Strategies for Structured Documents in RAG
No chunking strategy wins every RAG workload. Put simple, Markdown, section-aware, and semantic chunkers behind one interface, preserve metadata, and choose by measured retrieval quality.
Artwork: Shoes by Vincent van Gogh. The Metropolitan Museum of Art · Public domain
Machine LearningWiring ONNX Runtime GenAI into a LangChain RAG Pipeline
A small LangChain adapter can run ONNX Runtime GenAI locally, but deployment performance depends on building and shipping the optimized Phi-3 graph instead of loading a raw export.
Artwork: Man in a Turban by Rembrandt. The Metropolitan Museum of Art · Public domain
Machine LearningChoosing an On-Prem SLM Inference Engine for Windows
ONNX Runtime GenAI was the practical default for offline Phi-3 inference on Windows: it led throughput and latency, stayed under five seconds, and avoided custom builds or Docker.