FreightIQX – Move LTL claims at the speed of freight

[prasadsatish (Satish Prasad)]

Submission type

Coded Agent with UiPath SDK

Name

Satish Prasad

Industry category in which use case would best fit in (Select up to 2 industries)

Logistic

Complexity level

Advanced

Summary (abstract)

LTL Claim Processing Coded Agent with UiPath SDK is an intelligent automation system designed to revolutionize Less-Than-Truckload (LTL) freight claim management. Traditional claims processing is slow, error-prone, and highly manual, leading to operational inefficiencies, customer dissatisfaction, and financial loss. This project addresses these challenges by integrating AI-driven reasoning agents, UiPath automation services, and a modern web portal into a unified, end-to-end claims platform.

The solution combines three major components:

ReAct Claims Agent (Python + LangGraph) — an autonomous AI agent that leverages reasoning-acting loops to analyze documents, validate claim data, detect anomalies, and make data-driven claim decisions.

LTL Claims Portal (React + TypeScript) — a user-friendly web interface enabling customers to submit, track, and interact with claims in real time.

Backend API (Node.js + UiPath SDK) — a middleware service that bridges the web portal, UiPath Orchestrator, and Data Services, ensuring secure and scalable workflow execution.

By integrating UiPath Document Understanding, Data Services, and Orchestrator, the system automates claim extraction, validation, and decision-making with minimal human intervention. It achieves up to 85% faster processing, 95% accuracy, and 60% cost reduction, while providing full transparency and auditability.

Detailed problem statement

The $90B+ U.S. Less-Than-Truckload (LTL) freight market faces growing pressure from tighter margins, increasing claim volumes, and fragmented manual workflows. Despite advances in logistics technology, over 80% of 3PLs still process claims manually — relying on spreadsheets, emails, and paper documentation (DAT Freight Focus 2023)
.

Every year, 5–10% of all LTL shipments encounter OSD issues (overages, shortages, damages), resulting in hundreds of millions of dollars in unresolved claims and delayed reimbursements. Filing a single claim can take 20–40 minutes, while managing hundreds consumes entire back-office teams, creating operational bottlenecks and inconsistent outcomes.

Additionally, freight theft and liability risks — exceeding £1 billion annually in the UK (RHA Insurance Services)
— amplify the need for robust, auditable, and automated claim management.

In summary, the LTL industry struggles with:

Manual, time-consuming processes that reduce productivity

High error and dispute rates due to human data handling

Limited visibility and inconsistent decision-making across carriers

Rising financial losses from unresolved, delayed, or fraudulent claims

There is a clear need for an intelligent, end-to-end automated claims platform that leverages AI and UiPath automation to accelerate processing, ensure accuracy, and deliver real-time visibility for all stakeholders.

Detailed solution

    %% LTL Claims Processing Agent - Architecture Diagram
%% Shows the multi-agent system with specialized sub-agents

graph TB
    subgraph "Main Graph (LangGraph)"
        MAIN[Main Orchestration Flow<br/>11 Nodes + Conditional Routing]
    end
    
    subgraph "Specialized Sub-Agents"
        ORCH[Orchestrator Agent<br/>Model: GPT-4o<br/>Purpose: Planning & Coordination]
        DOC[Document Processor Agent<br/>Model: GPT-4o-mini<br/>Purpose: Document Download & Extraction]
        RISK[Risk Assessor Agent<br/>Model: GPT-4o-mini<br/>Purpose: Risk Analysis & Scoring]
        COMP[Compliance Validator Agent<br/>Model: GPT-4o-mini<br/>Purpose: Policy Validation]
    end
    
    subgraph "Decision Strategy"
        DEC[Hybrid Decision Strategy<br/>Model: GPT-4o<br/>LLM + Rule-Based Fallback]
    end
    
    subgraph "UiPath Services"
        DF[Data Fabric<br/>Entities API]
        IXP[Document Understanding<br/>IXP/DU API]
        CG[Context Grounding<br/>Knowledge Base]
        AC[Action Center<br/>Human-in-the-Loop]
        QUEUE[Queue Management<br/>Orchestrator Queues]
        BUCKET[Storage Buckets<br/>Document Storage]
    end
    
    subgraph "Memory System"
        MEM[Long-Term Memory<br/>SQLite/PostgreSQL<br/>Historical Context & Patterns]
    end
    
    %% Main Flow Connections
    MAIN -->|Create Plan| ORCH
    MAIN -->|Process Documents| DOC
    MAIN -->|Assess Risk| RISK
    MAIN -->|Validate Policy| COMP
    MAIN -->|Make Decision| DEC
    
    %% Sub-Agent to Service Connections
    ORCH -.->|Query Tools| CG
    DOC -->|Download| BUCKET
    DOC -->|Extract| IXP
    RISK -.->|Search Similar Claims| MEM
    COMP -->|Search Policies| CG
    COMP -->|Search Carriers| CG
    DEC -.->|Historical Context| MEM
    
    %% Main to Service Connections
    MAIN -->|Validate| DF
    MAIN -->|Escalate| AC
    MAIN -->|Update Status| QUEUE
    MAIN -->|Store Results| DF
    MAIN -->|Store Outcome| MEM
    
    %% Styling
    classDef agentClass fill:#e1f5ff,stroke:#0288d1,stroke-width:2px
    classDef serviceClass fill:#f3e5f5,stroke:#7b1fa2,stroke-width:2px
    classDef mainClass fill:#fff9c4,stroke:#f57f17,stroke-width:3px
    classDef memoryClass fill:#e8f5e9,stroke:#2e7d32,stroke-width:2px
    
    class ORCH,DOC,RISK,COMP,DEC agentClass
    class DF,IXP,CG,AC,QUEUE,BUCKET serviceClass
    class MAIN mainClass
    class MEM memoryClass

LTL Claims Processing Agent

Production-grade multi-agent system using LangGraph orchestration for intelligent claims processing

Technology Stack

• Core Framework: Python 3.10+, LangGraph, LangChain
• AI Models: GPT-4o (orchestration & decisions), GPT-4o-mini (specialized tasks)
• Integration: UiPath Python SDK v0.0.106+
• Memory: SQLite/PostgreSQL for long-term pattern learning
• State Management: Pydantic models with comprehensive validation

Architecture Highlights

• Multi-Agent System: 4 specialized sub-agents coordinated by LangGraph
• Orchestrator Agent (GPT-4o): Creates execution plans and coordinates workflow
• Document Processor Agent (GPT-4o-mini): Downloads and extracts document data
• Risk Assessor Agent (GPT-4o-mini): Calculates risk scores and identifies factors
• Compliance Validator Agent (GPT-4o-mini): Validates policy compliance
• Decision Strategy: Hybrid LLM + rule-based approach with fallback logic

State Graph: 11 nodes with conditional routing for complex workflows
Memory System: Long-term memory stores historical patterns for improved decisions

Key Capabilities

Document Understanding: Extracts data from BOLs, invoices, damage reports using UiPath IXP
Knowledge Search: Queries Context Grounding for policies, procedures, and precedents
Multi-Layer Validation: Data Fabric validation, document confidence checks, policy compliance
Risk Assessment: Weighted scoring algorithm with historical pattern analysis
� Human-ini-the-Loop: Action Center integration for low-confidence or high-risk claims
� * Intelligent Decisions*: Hybrid strategy combining LLM reasoning with rule-based fallbacks
� Auvdit Trail: Complete processing history with reasoning steps and tool usage
Learning System: Stores outcomes in memory for continuous improvement

Processing Workflow (11 Nodes)

Initialize Input - Load historical context from memory
Create Plan - Orchestrator generates execution plan
Validate Data - Query Data Fabric for claim/shipment validation
Download Documents - Process and extract document data
Assess Risk - Calculate risk score with weighted factors
Validate Policy - Check compliance against knowledge base
Evaluate Progress - Determine if human review needed
Escalate to Human - Create Action Center task (conditional)
Make Decision - Hybrid LLM + rule-based decision
Update Systems - Update queue transaction and Data Fabric
Finalize Output - Store in memory and build response

LTL Claims Web Portal

Modern, responsive React application for seamless claim submission and tracking

Technology Stack
Frontend Framework: React 18 with TypeScript
UI Library: Material-UI (MUI) v5 with custom theme
Styling: Tailwind CSS + MUI components
Build Tool: Vite for fast development and optimized builds

Narrated video link (sample: https://bit.ly/4pvuNEL)

FreightIQX – Move LTL claims at the speed of freight

Expected impact of this automation

Key Benefits

85% Faster Processing: Claims resolved in hours, not days

95% Accuracy: AI validation minimizes errors

60% Cost Reduction: Automation reduces manual workload

Full Auditability: Every decision is logged and traceable

Seamless Integration: Built on UiPath’s cloud ecosystem with scalable architecture

This solution transforms traditional, manual claim workflows into a smart, autonomous, and auditable system, empowering 3PLs and carriers to process claims faster, cut costs, and improve customer trust.

    %% LTL Claims Processing Agent - Complete Workflow Diagram
%% This diagram reflects the actual implementation in main.py

graph TB
    START([Start: Claim Input]) --> INIT[Initialize Input Node]
    
    %% Initialize Input Node
    INIT --> INIT_VALIDATE{Validate Input<br/>Fields}
    INIT_VALIDATE -->|Valid| INIT_MEMORY{Long-term<br/>Memory Enabled?}
    INIT_VALIDATE -->|Invalid| ERROR_INIT[Log Validation Errors]
    ERROR_INIT --> END_ERROR([End: Validation Failed])
    
    %% Memory Loading
    INIT_MEMORY -->|Yes| LOAD_MEMORY[Load Historical Context<br/>- Similar Claims<br/>- Decision Patterns]
    INIT_MEMORY -->|No| CREATE_PLAN
    LOAD_MEMORY --> CREATE_PLAN[Create Plan Node]
    
    %% Plan Creation
    CREATE_PLAN --> PLAN_AGENT[Orchestrator Agent<br/>GPT-4o]
    PLAN_AGENT --> PLAN_TOOLS{Use Tools?}
    PLAN_TOOLS -->|Yes| PLAN_TOOLS_EXEC[Execute Planning Tools]
    PLAN_TOOLS_EXEC --> PLAN_RESULT
    PLAN_TOOLS -->|No| PLAN_RESULT[Generate Execution Plan]
    PLAN_RESULT --> VALIDATE_DATA
    
    %% Data Validation
    VALIDATE_DATA[Validate Data Node] --> DF_QUERY[Query Data Fabric<br/>- Validate Claim ID<br/>- Validate Shipment ID]
    DF_QUERY --> DF_RESULT{Data Found?}
    DF_RESULT -->|Yes| DF_ENRICH[Enrich State with<br/>Data Fabric Info]
    DF_RESULT -->|No| DF_ERROR[Add Validation Error]
    DF_ENRICH --> DOWNLOAD_DOCS
    DF_ERROR --> DOWNLOAD_DOCS
    
    %% Document Processing
    DOWNLOAD_DOCS[Download Documents Node] --> DOCS_CHECK{Documents<br/>Available?}
    DOCS_CHECK -->|No| ASSESS_RISK
    DOCS_CHECK -->|Yes| DOC_AGENT[Document Processor Agent<br/>GPT-4o-mini]
    
    DOC_AGENT --> DOC_DOWNLOAD[Download from Storage<br/>- Shipping Documents<br/>- Damage Evidence]
    DOC_DOWNLOAD --> DOC_EXTRACT[Extract Data via IXP<br/>- Document Understanding<br/>- Confidence Scores]
    DOC_EXTRACT --> DOC_CONFIDENCE{Low Confidence<br/>Fields?}
    DOC_CONFIDENCE -->|Yes| DOC_FLAG[Flag for Review]
    DOC_CONFIDENCE -->|No| DOC_COMPLETE
    DOC_FLAG --> DOC_COMPLETE[Store Extracted Data]
    DOC_COMPLETE --> ASSESS_RISK
    
    %% Risk Assessment
    ASSESS_RISK[Assess Risk Node] --> RISK_FACTORS[Collect Risk Factors<br/>- High Amount<br/>- Claim Type<br/>- Low Confidence<br/>- Missing Docs<br/>- Policy Violations]
    RISK_FACTORS --> RISK_CALC[Calculate Risk Score<br/>Weighted Algorithm]
    RISK_CALC --> RISK_AGENT[Risk Assessor Agent<br/>GPT-4o-mini]
    RISK_AGENT --> RISK_REASONING[Generate Risk Reasoning]
    RISK_REASONING --> RISK_LEVEL{Risk Level?}
    RISK_LEVEL -->|Low| RISK_LOW[Risk: Low]
    RISK_LEVEL -->|Medium| RISK_MED[Risk: Medium]
    RISK_LEVEL -->|High| RISK_HIGH[Risk: High]
    RISK_LOW --> VALIDATE_POLICY
    RISK_MED --> VALIDATE_POLICY
    RISK_HIGH --> VALIDATE_POLICY
    
    %% Policy Validation
    VALIDATE_POLICY[Validate Policy Node] --> COMP_AGENT[Compliance Validator Agent<br/>GPT-4o-mini]
    COMP_AGENT --> COMP_SEARCH[Search Knowledge Base<br/>- Claims Policies<br/>- Carrier Liability<br/>- Procedures]
    COMP_SEARCH --> COMP_CHECK[Check Violations<br/>- Amount Limits<br/>- Carrier Liability<br/>- Required Docs]
    COMP_CHECK --> COMP_RESULT{Violations<br/>Found?}
    COMP_RESULT -->|Yes| COMP_VIOLATIONS[Record Violations]
    COMP_RESULT -->|No| COMP_COMPLIANT[Mark Compliant]
    COMP_VIOLATIONS --> EVALUATE_PROGRESS
    COMP_COMPLIANT --> EVALUATE_PROGRESS
    
    %% Progress Evaluation
    EVALUATE_PROGRESS[Evaluate Progress Node] --> EVAL_CONFIDENCE{Confidence<br/>< Threshold?}
    EVAL_CONFIDENCE -->|Yes| EVAL_ESCALATE[Flag for Review]
    EVAL_CONFIDENCE -->|No| EVAL_RISK{Risk Level<br/>High?}
    EVAL_RISK -->|Yes| EVAL_ESCALATE
    EVAL_RISK -->|No| EVAL_VIOLATIONS{Policy<br/>Violations?}
    EVAL_VIOLATIONS -->|Yes| EVAL_ESCALATE
    EVAL_VIOLATIONS -->|No| EVAL_ERRORS{Critical<br/>Errors?}
    EVAL_ERRORS -->|Yes| EVAL_ESCALATE
    EVAL_ERRORS -->|No| EVAL_CONTINUE[Continue to Decision]
    EVAL_ESCALATE --> ESCALATE_CHECK
    EVAL_CONTINUE --> MAKE_DECISION
    
    %% Human Escalation
    ESCALATE_CHECK{Action Center<br/>Enabled?}
    ESCALATE_CHECK -->|No| ESCALATE_SKIP[Skip Escalation]
    ESCALATE_CHECK -->|Yes| ESCALATE_TO_HUMAN[Escalate to Human Node]
    ESCALATE_SKIP --> MAKE_DECISION
    
    ESCALATE_TO_HUMAN --> AC_CREATE[Create Action Center Task<br/>- Claim Details<br/>- Risk Factors<br/>- Extracted Data]
    AC_CREATE --> AC_WAIT[Wait for Human Decision]
    AC_WAIT --> AC_DECISION{Human<br/>Decision?}
    AC_DECISION -->|Approved| AC_APPROVE[Set Decision: Approved]
    AC_DECISION -->|Denied| AC_DENY[Set Decision: Denied]
    AC_DECISION -->|Pending| AC_PENDING[Set Decision: Pending]
    AC_APPROVE --> MAKE_DECISION
    AC_DENY --> MAKE_DECISION
    AC_PENDING --> MAKE_DECISION
    
    %% Decision Making
    MAKE_DECISION[Make Decision Node] --> DECISION_LLM[Decision Strategy<br/>GPT-4o]
    DECISION_LLM --> DECISION_CONTEXT[Build Decision Context<br/>- Claim Data<br/>- Risk Assessment<br/>- Policy Compliance<br/>- Historical Context]
    DECISION_CONTEXT --> DECISION_INVOKE[Invoke LLM Decision]
    DECISION_INVOKE --> DECISION_PARSE[Parse Decision Response<br/>- Decision<br/>- Confidence<br/>- Reasoning]
    DECISION_PARSE --> DECISION_VALIDATE{Valid<br/>Decision?}
    DECISION_VALIDATE -->|No| DECISION_FALLBACK[Use Rule-Based Fallback]
    DECISION_VALIDATE -->|Yes| DECISION_RESULT
    DECISION_FALLBACK --> DECISION_RESULT[Store Decision]
    DECISION_RESULT --> UPDATE_SYSTEMS
    
    %% Update Systems
    UPDATE_SYSTEMS[Update Systems Node] --> UPDATE_QUEUE{Transaction<br/>Key Exists?}
    UPDATE_QUEUE -->|Yes| QUEUE_UPDATE[Update Queue Transaction<br/>- Status<br/>- Output Data<br/>- Error Messages]
    UPDATE_QUEUE -->|No| QUEUE_SKIP[Skip Queue Update]
    QUEUE_UPDATE --> UPDATE_DF
    QUEUE_SKIP --> UPDATE_DF
    
    UPDATE_DF[Update Data Fabric<br/>- Claim Status<br/>- Processing History<br/>- Decision Details] --> FINALIZE
    
    %% Finalize Output
    FINALIZE[Finalize Output Node] --> FINALIZE_MEMORY{Long-term<br/>Memory Enabled?}
    FINALIZE_MEMORY -->|Yes| STORE_MEMORY[Store Outcome in Memory<br/>- Decision<br/>- Confidence<br/>- Reasoning<br/>- Outcome]
    FINALIZE_MEMORY -->|No| BUILD_OUTPUT
    STORE_MEMORY --> BUILD_OUTPUT
    
    BUILD_OUTPUT[Build Output Response<br/>- Success Status<br/>- Decision<br/>- Confidence<br/>- Reasoning<br/>- Audit Trail] --> END_SUCCESS([End: Processing Complete])
    
    %% Styling
    classDef agentNode fill:#e1f5ff,stroke:#0288d1,stroke-width:2px
    classDef decisionNode fill:#fff9c4,stroke:#f57f17,stroke-width:2px
    classDef errorNode fill:#ffebee,stroke:#c62828,stroke-width:2px
    classDef successNode fill:#e8f5e9,stroke:#2e7d32,stroke-width:2px
    classDef toolNode fill:#f3e5f5,stroke:#7b1fa2,stroke-width:2px
    
    class PLAN_AGENT,DOC_AGENT,RISK_AGENT,COMP_AGENT,DECISION_LLM agentNode
    class INIT_VALIDATE,INIT_MEMORY,DOCS_CHECK,DOC_CONFIDENCE,RISK_LEVEL,COMP_RESULT,EVAL_CONFIDENCE,EVAL_RISK,EVAL_VIOLATIONS,EVAL_ERRORS,ESCALATE_CHECK,AC_DECISION,DECISION_VALIDATE,UPDATE_QUEUE,FINALIZE_MEMORY decisionNode
    class ERROR_INIT,DF_ERROR,DOC_FLAG,COMP_VIOLATIONS,EVAL_ESCALATE errorNode
    class END_SUCCESS,COMP_COMPLIANT,EVAL_CONTINUE successNode
    class PLAN_TOOLS_EXEC,DF_QUERY,DOC_DOWNLOAD,DOC_EXTRACT,COMP_SEARCH,AC_CREATE,QUEUE_UPDATE,UPDATE_DF,STORE_MEMORY toolNode

UiPath products used (select up to 4 items)

UiPath Action Center
UiPath Automation Cloud™
UiPath Data Service
UiPath Document Understanding™
UiPath IXP
UiPath Orchestrator
UiPath Studio Web

Automation Applications

TMS

Integration with external technologies

Open AI

TO-BE workflow/architecture diagram (file size up to 4 MB)

Other resources

GitHub -

GitHub - SATMAN778/FreightIQX-UiPathCoded-Agent-for-LTL-Claims

Additional Content

Google Drive: Sign-in

Backend Code added in Samples as PR - Added ltl-claims-agents in samples by SATMAN778 · Pull Request #255 · UiPath/uipath-langchain-python · GitHub

[loredana_ifrim]

Hi there, @prasadsatish builder,

It’s been a while . Thank you so much for being part of the Specialist Coded Agent Challenge. Your creativity, dedication, and automation skills truly blew us away!

Here’s what’s next:

Nov 5–16: Jury evaluation by @eusebiu.jecan1 & @Adrian_Tamas + community voting
Nov 17: Winners announced

Don’t forget the Community Choice Award, the best-voted project wins a $500 gift card + $60 UiPath Swag voucher! Voting is open till Nov 16, but remember that fresh accounts can’t vote (Level 1 access required, as we want to keep it fair and spam-free).

You’ve already won our admiration, now let’s see who takes home the big prizes .

GOOD LUCK ,

Loredana

[edevries (Peter Griffin)]

Thank you for sharing. Good inspiration for those of us working in the transportation and logistics industry. I think some of these agentic tools could be of good use, but we need use-cases and examples like this before management green lights investment into AI credits.