Web Interface Guide

Relevant source files

The following files were used as context for generating this wiki page:

• frontend/index.html
• frontend/app.js
• backend.py
• frontend/static.css
• pipeline.py
• README.md

Introduction

The Rosy Web Interface serves as the primary user-facing layer for the AI-powered tax document processing system. It provides a streamlined workflow for taxpayers to upload financial documents, review AI-extracted data, and generate completed tax forms. The interface is built as a Single Page Application (SPA) architecture, utilizing a vanilla JavaScript frontend that communicates with a Flask-based REST API.

The guide covers the components responsible for document ingestion, real-time status reporting, interactive data validation, and the finalization of Form 1040 documents.
Sources: README.md:9-16, backend.py:15-18

System Architecture & Data Flow

The web interface acts as the entry point for the processing pipeline. Data flows from the user's local machine through the backend orchestration layer and back to the user as a processed PDF.

The diagram above illustrates the request-response lifecycle between the web interface and the backend processing modules.
Sources: frontend/app.js:1-60, backend.py:65-95, pipeline.py:44-55

Core Components

Document Upload Interface

The upload section allows users to submit multiple files (PDF, TXT, PNG, JPG) simultaneously. Users must also specify their filing status and optional withholding amounts which are used to initialize the tax calculation engine.

Element	Type	Description
filing_status	Select	Options: Single, Married. Used for standard deduction selection.
withholding	Input (Number)	Optional manual withholding override.
files	Input (File)	Multiple file selection for W-2 or 1099 documents.
upload-form	Form	Triggers the async processing sequence.

Sources: frontend/index.html:13-28, frontend/app.js:6-21

Interactive Review & Editing

Upon successful processing, the interface dynamically generates a review UI. This allows users to inspect the AI's "confidence" scores and manually correct any misidentified fields before the final tax calculation.

• Per-File Parsing: Each uploaded document is displayed in a dedicated box showing the detected document type (e.g., W-2, 1099) and its specific fields.
• Real-time Updates: The Save changes button updates the client-side state and triggers a re-render of the "Aggregated Preview" section.
• Field Mapping: Input fields are dynamically generated based on the keys returned in the pf.fields object from the backend.

Sources: frontend/app.js:46-85, pipeline.py:17-38

Finalization and PDF Generation

The final stage of the web interface workflow is the "Finalize & Download" action. This sends the validated and potentially edited data back to the server to produce the final document.

This sequence ensures that the final PDF reflects any manual corrections made by the user during the review phase.
Sources: frontend/app.js:100-116, backend.py:114-153

API Integration

The frontend interacts with the following primary endpoints defined in the backend:

1. File Processing (POST /upload)

• Payload: multipart/form-data containing files and metadata.
• Logic: Sanitizes filenames and passes them to the run_pipeline_on_paths function.
• Response: Returns a JSON object containing doc_type, confidence, per_file results, and an initial tax_estimate.
  Sources: backend.py:65-101, frontend/app.js:25-39

2. PDF Finalization (POST /finalize)

• Payload: application/json containing the edited per_file field data.
• Logic: Aggregates the edited values, recomputes the final tax liability using compute_tax_estimate, and fills the PDF template.
• Response: Binary stream of the generated PDF file.
  Sources: backend.py:114-153, frontend/app.js:103-107

Styling and Layout

The interface utilizes a clean, centered layout optimized for document management.

• Max Width: 900px for readability.
• Visual Feedback: The #status element provides asynchronous updates (e.g., "Uploading...", "Processing complete") using aria-live="polite" for accessibility.
• Code Previews: Aggregated data and tax estimates are rendered within <pre> blocks for technical transparency during the prototype phase.
  Sources: frontend/static.css:1-6, frontend/index.html:30, frontend/app.js:43-52

Conclusion

The Web Interface Guide demonstrates a modular approach to tax document processing. By separating the initial extraction from the final generation, the system allows for a "human-in-the-loop" validation step, ensuring high accuracy in the final Form 1040 production. The interface handles the complexities of asynchronous file processing and binary data management to provide a seamless user experience.