Practical Examples

Three examples of increasing complexity: from the minimal RAG flow to a complete insurance agent with form, categorization, routing, suspension, and differentiated reports.

---

Example 1 — RAG Chat (Simple)

Objective: Answer user questions by drawing on company documents and the Knowledge Base. The simplest and most common flow.

Flow:

Begin → Retrieval → LLM → Message

Complete Configuration

Begin Node

Parameter	Value
mode	conversational
inputFields	(empty)

Retrieval Node (retrieval_1)

Parameter	Value
query	{{sys.query}}
topK	10
scoreThreshold	0.5
useReranking	true
companyDocs	true
knowledgeBase	true
legalSources	false

LLM Node (llm_1)

Parameter	Value
model	writer
temperature	0.3
maxTokens	2048

System Prompt:

You are a precise and professional corporate assistant. Always respond in English.
Base your answers EXCLUSIVELY on the documents provided in context.
If the requested information is not in the documents, say so explicitly: do not invent.
Always cite sources with numbered references [N].

User Prompt:

Context documents:
{{retrieval_1.formalized_content}}

User question:
{{sys.query}}

Message Node (message_1)

Parameter	Value
content	{{llm_1.content}}
format	markdown
showCitations	true

What You Learn

• How to use {{sys.query}} as a retrieval query.
• How to pass {{retrieval_1.formalized_content}} to the LLM prompt.
• The fundamental pattern Begin → Retrieval → LLM → Message.
• How the Message node handles citations and formatting.

---

Example 2 — Smart Router (Medium)

Objective: Handle different types of questions with different LLMs. Technical questions use the writer model for in-depth responses; simple questions use the planner model for speed; out-of-scope questions receive a static message.

Flow:

Begin → Categorize → [Technical] → LLM writer → Message technical response
                   → [Simple]    → LLM planner → Message quick response
                   → [Out of scope] → Static Message

Complete Configuration

Begin Node

Parameter	Value
mode	conversational

Categorize Node (categorize_1)

Parameter	Value
input	{{sys.query}}
minConfidence	0.65

Categories:

Name	Description	Goto
Technical	Questions about products, technical specifications, procedures, manuals, installations, configurations, technical faults	retrieval_technical
Simple	General questions, greetings, short clarification requests, questions about basic services	llm_quick
OutOfScope	Questions not relevant to company activities, personal topics, prohibited content	message_out_of_scope

Technical Branch

Retrieval Node (retrieval_technical):

Parameter	Value
query	{{sys.query}}
topK	15
useReranking	true
companyDocs	true
knowledgeBase	true

LLM writer Node (llm_technical):

Parameter	Value
model	writer
temperature	0.2
maxTokens	3000

System Prompt:

You are a corporate technical expert. Provide detailed, precise, and structured answers.
Use bullet points and headings where useful. Cite sources with [N].
The question was classified as TECHNICAL (confidence: {{categorize_1.confidence}}).

User Prompt:

Available technical documentation:
{{retrieval_technical.formalized_content}}

Technical question:
{{sys.query}}

Message Node (message_technical):

Parameter	Value
content	{{llm_technical.content}}
format	markdown
showCitations	true

Simple Branch

LLM planner Node (llm_quick):

Parameter	Value
model	planner
temperature	0.5
maxTokens	512

System Prompt:

You are a friendly corporate assistant. Respond concisely and directly.

User Prompt:

{{sys.query}}

Message Node (message_quick):

Parameter	Value
content	{{llm_quick.content}}
format	markdown
showCitations	false

Out of Scope Branch

Message Node (message_out_of_scope):

Parameter	Value
content	This question falls outside the scope of the assistant. You can ask me about products, technical procedures, or company services.
format	plain
showCitations	false

What You Learn

• How to use Categorize for semantic routing without rigid conditions.
• How to create parallel branches with different LLMs for efficiency and quality.
• How to use {{categorize_1.confidence}} in the prompt for transparency.
• How to terminate branches with static or dynamic messages.

---

Example 3 — Insurance Claim Analysis (Complex)

Objective: Manage the entire evaluation pipeline for an insurance claim. The flow collects initial data, categorizes the claim type, retrieves applicable policies and regulations, analyzes coverage, and — if coverage is not confirmed — requests supplementary documentation from the user before generating the final report.

Complete flow:

Begin (form)
  → Categorize (claim type)
    → VariableAssigner (prepare query)
      → Retrieval (policies + regulations)
        → LLM Analysis (coverage assessment)
          → Switch (COVERAGE CONFIRMED?)
            → [YES] LLM Positive Report → Message Positive Outcome
            → [NO] UserFillUp (supplementary documents)
                     → Supplementary Retrieval
                       → LLM Negative Report → Message Negative Outcome

Complete Configuration

Begin Node

Parameter	Value
mode	task

Input Fields:

Name	Label	Type	Required
claimType	Claim Type	select [Auto, Life, Home, Liability, Other]	Yes
claimDate	Date of Incident	date	Yes
description	Description of Event	text	Yes
policyNumber	Policy Number	string	Yes
estimatedAmount	Estimated Damage Amount (€)	number	No

Categorize Node (categorize_1)

Parameter	Value
input	{{sys.inputs.claimType}}: {{sys.inputs.description}}
minConfidence	0.70

Categories:

Name	Description	Goto
Auto	Road accidents, vehicle damage, car theft, auto liability	varassign_query
Life	Death, permanent disability, serious illness, injuries	varassign_query
Home	Property damage, fire, flood, home theft, home liability	varassign_query
Liability	Civil liability toward third parties, damage to persons or property caused by the insured	varassign_query
Other	Claims not classifiable in the above categories	varassign_query

(All categories converge on the next node; the value {{categorize_1.category}} is used in the query)

VariableAssigner Node (varassign_query)

assignments:
  - name: searchQuery
    type: concat
    value: "Claim {{categorize_1.category}} policy {{sys.inputs.policyNumber}} coverage conditions exclusions {{sys.inputs.description}}"
  - name: claimSummary
    type: concat
    value: |
      CLAIM TYPE: {{sys.inputs.claimType}} (category: {{categorize_1.category}})
      DATE OF INCIDENT: {{sys.inputs.claimDate}}
      POLICY NUMBER: {{sys.inputs.policyNumber}}
      ESTIMATED AMOUNT: {{sys.inputs.estimatedAmount}} EUR
      DESCRIPTION: {{sys.inputs.description}}

Main Retrieval Node (retrieval_policies)

Parameter	Value
query	{{varassign_query.searchQuery}}
topK	20
scoreThreshold	0.45
useReranking	true
companyDocs	true
knowledgeBase	true
legalSources	true

Coverage Analysis LLM Node (llm_analysis)

Parameter	Value
model	writer
temperature	0.1
maxTokens	3000
jsonMode	false

System Prompt:

You are a senior insurance assessor with 20 years of experience.
Analyze the provided documentation and determine whether the claim is covered by the policy.

CRITICAL RULES:
1. Base the assessment EXCLUSIVELY on the provided documents. Do not assume undocumented coverage.
2. Always cite contractual and regulatory references with [N].
3. ALWAYS conclude your analysis with one of these exact phrases:
   - "VERDICT: COVERAGE CONFIRMED" if the claim falls within the policy coverage
   - "VERDICT: COVERAGE NOT CONFIRMED" if there is insufficient evidence of coverage or applicable exclusions apply
4. After the verdict, list any additional documentation required (if any).

User Prompt:

CLAIM DATA:
{{varassign_query.claimSummary}}

POLICY AND REGULATORY DOCUMENTATION:
{{retrieval_policies.formalized_content}}

Perform the complete assessment following the operational rules.

Switch Node (switch_coverage)

Parameter	Value
elseGoto	userfillup_supplementary

Conditions:

Variable	Operator	Value	Goto
{{llm_analysis.content}}	contains	COVERAGE CONFIRMED	llm_report_positive

Positive Branch — LLM Report (llm_report_positive)

Parameter	Value
model	writer
temperature	0.2
maxTokens	2500

System Prompt:

You are an insurance communication expert. Draft professional and clear reports.

User Prompt:

Based on the following assessment, draft an official claim acceptance report
in formal English, structured with sections: Summary, Documentation Reviewed,
Coverage Assessment, Next Steps.

CLAIM DATA:
{{varassign_query.claimSummary}}

ASSESSMENT:
{{llm_analysis.content}}

Positive Outcome Message Node (message_positive)

Parameter	Value
content	{{llm_report_positive.content}}
format	markdown
showCitations	true

Negative Branch — UserFillUp (userfillup_supplementary)

Parameter	Value
message	The initial analysis did not identify certain coverage. To proceed with the assessment, we ask you to provide the supplementary documentation indicated below.

Fields:

Name	Label	Type	Required
damagePictures	Photographs of the damage	file	Yes
appraisal	Technical appraisal or damage estimate	file	No
ownershipDocument	Proof of ownership of the insured property	file	No
additionalNotes	Additional notes or clarifications	text	No

Tips: Accepted formats: PDF, JPG, PNG, DOCX. Maximum size per file: 20MB.

Supplementary Retrieval (retrieval_supplementary)

Parameter	Value
query	{{varassign_query.searchQuery}} exclusions limitations deductible maximum limits
topK	15
companyDocs	true
knowledgeBase	true
legalSources	true

Negative Report LLM (llm_report_negative)

Parameter	Value
model	writer
temperature	0.1
maxTokens	3000

System Prompt:

You are an insurance communication expert. Draft professional, transparent, and precise reports.
The report must be clear about the reasons for denial or the supplementary request,
and accurately indicate the documentation required to proceed.

User Prompt:

Draft an official supplementary documentation request report for the following claim.
Structure the report with sections: Claim Summary, Preliminary Analysis,
Reason for Supplementary Request, Documentation Received, Next Steps.

CLAIM DATA:
{{varassign_query.claimSummary}}

INITIAL ASSESSMENT:
{{llm_analysis.content}}

SUPPLEMENTARY DOCUMENTATION RECEIVED:
{{userfillup_supplementary.inputs.additionalNotes}}

ADDITIONAL REGULATIONS AND CONDITIONS:
{{retrieval_supplementary.formalized_content}}

Negative Outcome Message Node (message_negative)

Parameter	Value
content	{{llm_report_negative.content}}
format	markdown
showCitations	true

What You Learn

• How to build a structured form with Begin in task mode.
• How to use Categorize for semantic routing with multiple categories.
• How VariableAssigner prepares composite data for subsequent nodes.
• How LLM with a rigorous System Prompt produces predictable output for Switch.
• How Switch routes based on the textual content of the LLM response.
• How UserFillUp suspends the flow and collects additional documents.
• How a second Retrieval enriches context after user input.
• How to differentiate the tone and structure of reports for positive and negative outcomes.

---

Queria v3.1.2 -- Canvas Agent Builder