AI ENGINEERING LAB

Choose the pattern.
Prove it can operate.

Five browser-based tools for architecture selection, autonomy, evaluations, MCP security and workload economics.

REVIEWED SEP 06, 2026PRACTITIONER-LED

XTIANZ / LAB

task      → classify
evidence  → retrieve
action    → constrain
result    → verify
risk      → gate
release   → observe

BROWSER TOOLS · NO API KEY

TOOL 01

Architecture pattern recommender

Use the least complex pattern that can meet the task, evidence and control requirements.

How fresh must the answer be?Does the task require system actions?Can success be automatically verified?How sensitive is the workflow?How much path variation is necessary?Is human approval available?

RECOMMENDED PATTERN

RAG + deterministic workflow

Ground changing knowledge and keep orchestration explicit.

AUTONOMY · LOWHUMAN GATE · OPTIONAL

TOOL 02

Agent deployment decision tree

Process mostly deterministic?Result automatically verifiable?Actions reversible?High-impact or sensitive?Human approval before actions?Failure cost?

RECOMMENDATION

Start with a deterministic workflow

Use rules first; add AI only where language interpretation is necessary.

TOOL 03

Evaluation-plan generator

TaskExpected outcomeKnown failure modesGround truthLatency targetCost target

EVALUATION OUTLINE

Describe the task to generate a release-ready evaluation outline.

TOOL 04

MCP threat-model builder

CONTROL CHECKLIST

Authenticated actor and server identity
Least privilege per tool
Structured logging and revocation path

TOOL 05

AI workload economics calculator

Tasks / monthModel calls / taskAvg tokens / callModel $ / 1M tokensRetry / failure %Human review %Minutes / reviewReviewer $ / hour

PLANNING ESTIMATEMODEL USAGE$486 / moHUMAN REVIEW$583 / moTOTAL$1,069 / mo

Planning aid only; provider pricing, caching, input/output ratios and infrastructure costs vary.

Choose the pattern.Prove it can operate.