You did it!¶
Congratulations!
You've built a complete code-first loan underwriting pipeline — from application intake through AI credit assessment to human review and final decision, all orchestrated by Maestro.
What you built¶
Four components working together as a single automated pipeline. The RPA workflow reads a CSV of applicants, submits each one to UiBank, and starts a Maestro Flow process for every accepted application. Inside the flow, the Python agent runs a full credit assessment using the UiBank API and surfaces a credit score, risk band, and assessment summary. Low-risk applications are auto-approved and the flow ends. Medium and high-risk ones pause and open the React dashboard in Action Center — the loan officer reviews the AI assessment and submits a decision. The flow captures that decision and outputs a final result. Back in the RPA, results are written to results.csv and uploaded to a storage bucket.
| Component | Role |
|---|---|
| Coded Agent (Python / LangGraph) | Receives a loan_id, calls the UiBank API across three tools, computes financial metrics, and uses the LLM to produce a credit score, risk band, and assessment summary |
| Coded Action App (React / TypeScript) | Loan officer dashboard — receives all twelve inputs from Maestro and captures an Approve or Reject decision |
| Maestro Flow | Orchestrates the end-to-end flow — calls the agent, routes by risk band, opens the app for human review, and outputs final_decision |
| Studio RPA Workflow | Reads loan-applications.csv, submits each row to UiBank, starts the Maestro Flow per accepted application, updates results.csv, and uploads it to a storage bucket |
Every component was generated by your coding agent using UiPath skills and deployed with the uip CLI.
Next steps¶
1. Trigger a live end-to-end run¶
In Orchestrator, navigate to your CodingAgentsILT folder. Open {YourName}LoanApplicationDispatcher and start a run. Watch the RPA submit applications, start the Maestro Flow, and route results back. For Medium and High-risk applications, open Action Center and complete the human review task — the loan officer dashboard should appear with all fields pre-filled.
2. Review the results¶
After a complete run, open results.csv. Each row should show either Loan Approved or Loan Rejected as the final status. Cross-reference with the Maestro execution trace in Orchestrator to confirm the routing matched the agent's risk band.
Keep iterating¶
Improve the underwriting logic
The LLM currently interprets three financial metrics holistically. Try refining the system prompt to give more weight to the loan-to-market ratio, or to treat young applicants with high income differently from older ones with moderate income. Retest with the same three sample profiles and see how the risk bands shift.
Add an auto-rejection path for extreme cases
Right now all Medium and High-risk applications go to a loan officer. Add a condition in the Maestro Flow: if credit_score < 20, auto-reject immediately. This cuts the officer's queue for clear-cut cases.
Extend the React app
Add a section to the dashboard that shows the three financial metrics as a mini chart — a bar for DTI, a gauge for credit score, a comparison line for the loan-to-market ratio. Give the loan officer a richer view of the numbers behind the AI recommendation.
Connect the rejection path to a notification
After a rejection is recorded, add an Integration Service step to send an email to the applicant. One extra node in the Maestro Flow after the rejected end state — a single connector call.
Learn more¶
| Resource | Description |
|---|---|
| UiPath RPA documentation | Studio XAML workflows, activity libraries, and Orchestrator publishing |
| UiPath Coded Agents documentation | Python agent framework — LangGraph integration, tool definitions, UiPath AI Trust Layer |
| UiPath Coded Apps documentation | React/TypeScript app development, action-schema.json, input/outcome patterns |
| UiPath Maestro Flow documentation | Flow JSON structure, node types, condition expressions, and variable mappings |
| LangGraph documentation | Graph-based agent framework — nodes, edges, tool calling, and state management |
UiPath CLI (uip) reference |
Full command reference for scaffolding, running, packing, and publishing |
Share your feedback
Have suggestions on the Coding Agents ILT content? We'd love to hear from you — reach out at madalina.popescu@uipath.com.