Blog
Engineering notes
· From the Lab
The SLM Governance Gap: When Small Models Lie About What They Did
A 1.5B model confidently reported deleting calendar events it never touched. What building a personal agent revealed about hallucinated actions vs hallucinated facts.
SLMGovernanceTool CallingArchitectureProduction
· From the Lab
Tool Hints vs Model Size: What Actually Makes SLMs Call Tools Correctly
Enum constraints on tool descriptions improved a 1.5B model's accuracy more than upgrading to 7B. Findings from building MCP servers with Ollama.
SLMMCPTool CallingArchitectureEdge AI
· From the Lab
PII Detection Without Cloud AI — Local SLM + Layered Architecture
Most PII detection is pattern matching — deterministic, sub-millisecond, and free. A layered architecture handles 98% without a model. A local SLM covers the remaining 2% without cloud tokens or data leaving your network.
ArchitecturePIIGDPRProductionSLMHealthcareFinanceTelecom