What you can buildwith what you learn here.
Every tutorial on STOE maps to a real architecture decision. Here's where those patterns show up in production systems — and how Chronicle ties them together as a reference.
Chronicle: AI meets real constraints
Chronicle is an end-to-end AI travel planning system built to demonstrate every concept on this site working together — RAG retrieval, multi-agent orchestration, constraint solving, risk scoring, prompt caching, and observability with Langfuse. It's the reference app we built while learning this stack.
3 per request
Itinerary candidates
74%
Cache hit rate
0.32
Avg. risk score
91%
Policy pass rate
Six domains. One shared stack.
The skills from each lab here map directly to production architecture decisions across industries.
AI Travel Planning — Chronicle
Chronicle is our reference architecture for constrained AI planning. It combines RAG-grounded destination knowledge with deterministic constraint checks, traveler profiles, risk scoring, and policy approvals — all driven by Claude and Qdrant.
Fraud Detection with Explainable AI
SHAP values make fraud model decisions auditable — every flagged transaction comes with a ranked list of contributing features. This is the pattern regulators increasingly expect in financial AI.
Bias-Aware Hiring Systems
Screening models that surface demographic parity gaps and equal-opportunity scores in real time. Understanding these trade-offs is the difference between fair AI and a compliance liability.
RAG-Powered Knowledge Assistants
Internal knowledge bases, policy retrieval, and document Q&A — all grounded in your own data through vector search. The same retrieval pattern that powers Chronicle's destination intelligence.
Explainable Clinical Decision Support
Diagnostic support models need to show their reasoning — a prediction without an explanation is not usable in clinical settings. Explainability is not a nice-to-have here; it's a safety requirement.
Agentic Workflow Automation
Multi-agent systems that break complex tasks into specialized sub-agents — planner, validator, risk engine, approver — with human-in-the-loop gates at the right points. This is the architecture pattern Chronicle demonstrates end-to-end.
Common patterns across every use case
These aren't separate skills. They're the same five patterns applied in different domains.
Retrieval
RAG + vector search
Reasoning
LLM + agent planning
Evaluation
ML metrics + SHAP
Constraints
Rules + policy checks
Explainability
Transparency + audit
Start with the underlying skills.
Every use case above is built from the same interactive labs. Pick the one that matches where you want to go.