Agent Builder use cases

Real-world examples of what you can build with FlowX.AI Agent Builder.

Preview

Agent Builder is currently in preview and may change before general availability.

Agent Builder supports a wide range of AI-powered automation scenarios. Below are the primary categories with examples drawn from production implementations.

Document processing and extraction

Extract structured data from unstructured documents — invoices, contracts, shipping documents, financial statements — using TEXT_EXTRACTION nodes with tailored prompts per document type.

Logistics document processing

An email-triggered pipeline processes carrier emails and their attachments:

• Classify documents by type (BOL, invoice, rate confirmation, lumper receipt, fuel receipt, and more)
• Extract structured fields from each document type using specialized TEXT_EXTRACTION nodes
• Reconcile extracted data against a Transportation Management System (TMS)
• Surface exceptions for human review via task management

AI nodes used: TEXT_EXTRACTION (12 document types), TEXT_UNDERSTANDING (12 comparison workflows), TEXT_GENERATION (email summary)

Patterns involved: Fan-out extraction, AI comparison

Mortgage document validation

Process mortgage application packages by extracting product data from bank documentation:

• Extract structured product details (interest rates, fees, insurance requirements) from bank product sheets
• Weight income and debt values per bank-specific rules using AI extraction
• Validate legal/regulatory eligibility using AI understanding

AI nodes used: TEXT_EXTRACTION (income/debt/product extraction), TEXT_UNDERSTANDING (eligibility checks)

Patterns involved: Hybrid AI + business rules, Fan-out extraction

---

Conversational AI

Build chat-based assistants that understand user intent, maintain conversation context, and provide personalized responses grounded in your data.

Mortgage advisor chatbot

A conversational assistant that guides users through mortgage product selection:

• Detect intent — classify user messages as greetings, product inquiries, data input, or other
• Route to specialized handlers (small talk, personalized offers, knowledge base Q&A)
• Generate personalized mortgage consultant reports with comparative cost tables
• Maintain conversation history across sessions

AI nodes used: TEXT_UNDERSTANDING (intent detection), TEXT_GENERATION (responses, reports), CUSTOM_AGENT (knowledge base RAG)

Patterns involved: Intent classification, Knowledge base RAG, Session state

---

AI-augmented decision making

Combine AI capabilities with deterministic business logic for auditable, explainable decision-making.

Product eligibility and scoring

Evaluate financial product eligibility using a pipeline that mixes AI and business rules:

• AI extracts income weights and debt factors from bank-specific rule documents
• Business rules compute PMT, DTI ratios, maximum loan amounts, and currency conversions
• AI filters products by qualitative criteria (loan type, sustainability features) with fuzzy matching
• Business rules normalize scores and rank the top products
• AI generates a professional consultant report with the final recommendations

AI nodes used: TEXT_EXTRACTION, TEXT_UNDERSTANDING, TEXT_TRANSFORMATION, TEXT_GENERATION

Patterns involved: Hybrid AI + business rules

Document reconciliation

Compare AI-extracted document data against system-of-record values:

• Compare field-by-field with structured exception reporting
• Generate match rates and confidence scores per document
• Flag mismatches, missing fields, and derived value discrepancies
• Route exceptions to human reviewers based on severity

AI nodes used: TEXT_UNDERSTANDING (comparison agents per document type)

Patterns involved: AI comparison and reconciliation

---

Email automation

Process incoming emails and their attachments automatically using email triggers and AI workflows.

Carrier email processing

Monitor an email inbox for incoming carrier communications:

• Trigger workflows automatically when emails arrive (IMAP integration)
• Summarize email content and classify intent (invoice submission, payment inquiry, dispute)
• Extract and process all attachments through the document pipeline
• Look up related orders in external systems (TMS) by BOL number
• Notify relevant staff of exceptions via email templates

AI nodes used: TEXT_GENERATION (email summary), TEXT_EXTRACTION (attachments), TEXT_UNDERSTANDING (comparison)

Data sources: Email Trigger, REST API (TMS), FlowX Database

---

Building your own use case

1. Identify the AI tasks

   Map which steps in your process need AI (extraction, classification, generation, comparison) versus deterministic logic (calculations, routing, validation).

2. Choose your patterns

   Select from the AI patterns that match your needs. Most apps combine 2-4 patterns.

3. Design the data flow

   Define your data sources, data model, and how data moves between processes, workflows, and AI nodes.

4. Build incrementally

   Start with a single AI node (e.g., one document type extraction), validate it works, then expand to the full pipeline.

5. Add human review

   Use task management views for exception handling and quality assurance, especially for high-stakes decisions.

Related resources

Tutorials

End-to-end build guides based on production apps

AI patterns

Reusable architectural patterns

Node types

Detailed node reference

Using agents

Deploy agents in your apps