Reference patterns
Examples
Runnable patterns with live demos, case studies, and architecture notes for review meetings — not a product you buy here.
Worked examples and decision guides you can run in reviews and pilots — an evidence base from The Ops Toolbox.
25
Patterns
Across 4 stacks
25
Live demos
Runnable interactive panels on this site
Pattern picker
Pick a profile close to your initiative. See representative examples and guides.
Policy or compliance answers that must cite approved sources.
Browse by pattern(Assistants, document Q&A, governance, platform, discovery)
Agents & copilots
Streaming tools, escalation, and human oversight.
- Operational Copilot with Streaming Tools
Operations staff get live help with customer lookup and ticket creation — with escalation when needed.
- Multi-model routing via unified gateway
Send the same request to different models and compare speed and quality through one endpoint.
- Intent Routing on Amazon Bedrock
Route questions by intent first — policy, technical, or escalation — then answer in the right tone.
- Human-in-the-Loop Approval Gates
Supervisors approve or reject CRM updates and refunds before anything runs.
- Policy-as-Code Gates Before Tool Calls
Rules block or route risky actions before an assistant can change live systems.
Grounded Q&A
Retrieval, citations, and knowledge-bound answers.
- Enterprise RAG on Azure AI Foundry
Staff ask policy questions and get answers backed by approved documents.
- AWS-Native Q&A with Amazon Bedrock
In-account Q&A using your knowledge base when configured, or seed content for early pilots.
- Policy Q&A with seed RAG
Ask policy questions and get grounded answers from in-app documents — with citations.
- Production Observability for AI Routes
See latency, tokens, matched documents, and rough cost on every request — ready for SRE and finance.
- Ingestion, Chunking, and Deterministic Citations
Turn messy source text into stable chunks and citation references for document Q&A and audits.
Governance & safety
Safety scoring, orchestration, and policy gates.
- Responsible AI Gate with Azure Content Safety
Check prompts and responses for harmful content before anything reaches customers.
- Durable multi-step orchestration
A reliable multi-step workflow for transformation intake — from brief to recommendation.
- Agent Orchestration on Azure AI Foundry
Answer questions in three governed steps: find context, draft answer, check safety.
- Orchestrated Agents on AWS Bedrock
Multi-step Q&A on AWS — full agent mode when configured, honest fallback pipeline when not.
- Policy Q&A with seed RAG
Ask policy questions and get grounded answers from in-app documents — with citations.
- Prompt A/B Evaluation with Structured Rubrics
Compare two prompts on the same message, then score which is clearer, safer, and better fit.
- Intent Routing on Amazon Bedrock
Route questions by intent first — policy, technical, or escalation — then answer in the right tone.
- Azure OpenAI Chat Baseline
Straight chat without document search — a fair comparison point before adding Q&A or workflows.
- Batch ticket triage
Classify a batch of support messages into category, priority, and queue in one pass.
- Human-in-the-Loop Approval Gates
Supervisors approve or reject CRM updates and refunds before anything runs.
- Production Observability for AI Routes
See latency, tokens, matched documents, and rough cost on every request — ready for SRE and finance.
- Ingestion, Chunking, and Deterministic Citations
Turn messy source text into stable chunks and citation references for document Q&A and audits.
- Policy-as-Code Gates Before Tool Calls
Rules block or route risky actions before an assistant can change live systems.
- PII Redaction Pre-Processor
Strip obvious personal data locally before anything is sent to an external model.
- Conversation Replay + Regression Evals
Store golden conversations and score prompt changes before they ship.
- Incident Assistant: Severity → Runbook → Comms
A multi-step incident helper using seed runbooks: classify severity, pick runbook, draft first comms.
- Vendor Questionnaire Summariser (Security + AI)
Extract what risk teams actually need from vendor docs: retention, regions, subprocessors, red flags, and follow-ups.
- Token Budget Guardrails (Per Session)
Hard caps per session to prevent pilot spend surprises, block after budget is exhausted.
Platform & routing
Gateways, evals, workflows, and intent routing.
- Multi-model routing via unified gateway
Send the same request to different models and compare speed and quality through one endpoint.
- Durable multi-step orchestration
A reliable multi-step workflow for transformation intake — from brief to recommendation.
- Agent Orchestration on Azure AI Foundry
Answer questions in three governed steps: find context, draft answer, check safety.
- Orchestrated Agents on AWS Bedrock
Multi-step Q&A on AWS — full agent mode when configured, honest fallback pipeline when not.
- Prompt A/B Evaluation with Structured Rubrics
Compare two prompts on the same message, then score which is clearer, safer, and better fit.
- Intent Routing on Amazon Bedrock
Route questions by intent first — policy, technical, or escalation — then answer in the right tone.
- Batch ticket triage
Classify a batch of support messages into category, priority, and queue in one pass.
- Production Observability for AI Routes
See latency, tokens, matched documents, and rough cost on every request — ready for SRE and finance.
- Conversation Replay + Regression Evals
Store golden conversations and score prompt changes before they ship.
- Incident Assistant: Severity → Runbook → Comms
A multi-step incident helper using seed runbooks: classify severity, pick runbook, draft first comms.
- Batch Clustering + Theme Labels
Group dozens of feedback lines into themes you can route to teams, without building a data platform.
Discovery & structure
Workshop extraction and initiative shaping.
- Transformation Discovery with Claude Structured Output
Turn workshop notes into structured initiative risks, stakeholders, and systems.
- Meeting Notes to Action Items with Claude
Turn meeting notes into a summary, action items with owners, and open questions.
- Vendor Questionnaire Summariser (Security + AI)
Extract what risk teams actually need from vendor docs: retention, regions, subprocessors, red flags, and follow-ups.
Batch & routing
Classify and route many items without a chat UI.
- Intent Routing on Amazon Bedrock
Route questions by intent first — policy, technical, or escalation — then answer in the right tone.
- Batch ticket triage
Classify a batch of support messages into category, priority, and queue in one pass.
- Batch Clustering + Theme Labels
Group dozens of feedback lines into themes you can route to teams, without building a data platform.
14 examples
Operations staff get live help with customer lookup and ticket creation — with escalation when needed.
Staff ask questions and get live answers in the chat
Send the same request to different models and compare speed and quality through one endpoint.
Same question sent to two models for a side-by-side comparison
Staff ask policy questions and get answers backed by approved documents.
Question searches approved documents before the model drafts an answer
Check prompts and responses for harmful content before anything reaches customers.
Your prompt is screened for harmful content before generation
Answer questions in three governed steps: find context, draft answer, check safety.
Find relevant policy excerpts, then draft and safety-check the answer
Ask policy questions and get grounded answers from in-app documents — with citations.
Policy question answered from in-app documents with citations
Compare two prompts on the same message, then score which is clearer, safer, and better fit.
Two prompts tested on the same message, then scored against a rubric
Classify a batch of support messages into category, priority, and queue in one pass.
Batch of support messages classified into category, priority, and queue
Supervisors approve or reject CRM updates and refunds before anything runs.
Assistant proposes an action; supervisor approves or rejects first
See latency, tokens, matched documents, and rough cost on every request — ready for SRE and finance.
Every request returns latency, tokens, matches, and rough cost
Store golden conversations and score prompt changes before they ship.
Golden conversation replayed and scored before prompt changes ship
A multi-step incident helper using seed runbooks: classify severity, pick runbook, draft first comms.
Incident summary turned into severity, runbook, actions, and comms draft
Group dozens of feedback lines into themes you can route to teams, without building a data platform.
Feedback lines grouped into themes you can route to teams
Hard caps per session to prevent pilot spend surprises, block after budget is exhausted.
Session spend tracked and blocked when the pilot budget is exhausted