Posted inAI System Design
Day 93: SLOs for AI Features
Define SLOs for AI features across availability, faithfulness, and latency — pick which two you page on.
Posted inAI AWS System Design
Day 52: Blackboard vs Message Bus Orchestration
Prefer SQS/EventBridge envelopes with artifact hashes over a shared mutable doc when you need auditability between agents.
Posted inAI System Design
Day 92: Threat Model an AI Feature
Threat-model AI features with STRIDE across tools, RAG corpora, and prompt-injection surfaces.
Posted inAI System Design
Day 51: Multi-Agent Roles: Planner, Implementer, Critic
Contracts between Planner, Implementer, and Critic — separate prompts, tools, and write surfaces so multi-agent work is orchestration, not shared mutable prompt soup.
Posted inAI System Design
Day 91: Architecture Review: Draw the Box Diagram First
Architecture review starts with a box diagram: data flow, trust boundaries, and SLOs before any model choice.
Day 90: Project: Incident Timeline Summarizer
Incident timeline summarizers must ground drafts in artifacts; humans edit before anything posts to Slack.
Posted inAI Machine Learning
Day 89: On-Device and Edge Completions
On-device and edge completions win when privacy, latency, or offline matter — keep cloud as the escalation path.
Posted inAI System Design
Day 88: Simulation and Digital Twins for Tools
Give agents digital-twin / dry-run APIs before real mutate tools — simulate first, apply second.
Posted inAI System Design
Day 87: Workflow Mining From Logs
Mine workflows from traces to propose runbooks, then require human edit before any automated remediation.
Posted inAI Machine Learning
Day 86: Search and Ranking Copilots
Search and ranking copilots need offline ranking deltas and online interleaving — anecdotes are not eval.