Cookbooks
Cookbooks
Sixteen practical, build-along recipes that teach the CALIBER platform end-to-end — each backed by a versioned, installable system example with readiness checks and an explicit review-before-activation boundary.
How to read each cookbook.
- Sign in at
/caliber/loginasadmin/adminon a running stack. - Each cookbook has a flow diagram, a numbered UI walkthrough (the dark chip is the exact navigation path; “Fill in” lists every field; “Click” is the button to press), a UI mockup, an assets table, and quality gates.
- Every green link opens the copy-paste asset file in that cookbook's
assets/folder — paste its contents into the matching field.
The UI-first promise. Open Cookbooks to install any of the 16 recipes as a paused workflow and editable draft. Review readiness before activation.
- Datasets fill from real runs:
Observability → open a trace → Add to test set. - Deterministic logic uses Data Transform for decisions, JSON Schema, mapping, confidence, and fixtures; Python Code remains a bounded extension.
- The Prompt Playground is a live chat (it calls the model); prompt regression is scored on the prompt's
Runsstage, whileEvaluationsscores a Test Set with deterministic graders. - Judges are for LLM-graded criteria; deterministic checks are scorers or assertions.
- Aria (12–15) plans from one sentence in
Plans; you create each artifact in its own page.
Pick a recipe
Each card opens a self-contained walkthrough — flow diagram, step-by-step UI snapshots, copy-paste assets, and quality gates.
Foundations
01
Trustworthy Intake Classifier
Prompt fundamentals: stable JSON output + a regression gate that catches weak edits.
Starter20–30 min
Open cookbook →
02
Precision Skills
Deterministic triggering + portable packaging — a skill that fires only when it should.
Starter25–40 min
Open cookbook →
03
Policy-Safe Decision Tool
Deterministic business logic first, then approval-gated side effects, then an explanation layer.
Core40–60 min
Open cookbook →
05
Governed Tool Connectivity (MCP)
Connect an external system, discover its tools, and prove a blocked tool refuses to run.
Core20–35 min
Open cookbook →
Build
04
Document-to-JSON Pipeline
Turn Office documents into schema-valid JSON, with readable failures for bad inputs.
Core35–55 min
Open cookbook →
06
Grounded Knowledge Assistant
Cited answers, honest abstention on missing/conflicting evidence, and a measured retrieval lift.
Core40–60 min
Open cookbook →
07
Support Triage Copilot
The full-stack recipe: classify → gather evidence → cited reply → approval-gated escalation.
Advanced60–90 min
Open cookbook →
08
Incident Response Copilot
Separate facts from hypotheses, recommend the lowest-risk action, and gate rollbacks.
Advanced60–90 min
Open cookbook →
Operate
09
Self-Healing Workflows
An operator playbook: reproduce a failure, localize it in the Run Monitor, patch, and validate.
Core30–50 min
Open cookbook →
16
Production Observability & Triage
Turn live traces into a triaged review queue + a regression test set — the production monitoring loop.
Core30–45 min
Open cookbook →
Govern
10
Trustworthy Evaluation
Certify that an automated judge agrees with human reviewers before it can block a release.
Core45–70 min
Open cookbook →
11
Release Signoff Factory
Aggregate cross-cookbook evidence into a defensible go / no-go decision.
Core30–50 min
Open cookbook →
Aria — Autonomous
12
Aria: Evaluation Harness from Intent
One sentence → Aria plans a faithfulness judge + a test set; you create them in the UI.
Aria15–25 min
Open cookbook →
13
Aria: Human-Review Queue from Intent
One sentence → Aria plans a review queue with a safety / citation / tone label schema.
Aria15–25 min
Open cookbook →
14
Aria: Governance Starter Kit from Intent
The flagship: one sentence → a plan for a judge + test set + review queue together.
Aria20–30 min
Open cookbook →
15
Aria: Triage & Recalibrate Loop
One sentence → review the weak runs, then kick off an async workflow calibration.
Aria20–35 min
Open cookbook →