Lapor AI

Inspiration

Government websites often ask people to complete high-stakes processes through dense forms, unfamiliar terminology, and long sequences of pages. Indonesian tax filing through Coretax is one example. Taxpayers need help understanding what belongs in each section, but they also need to verify every value before filing.

I built LaporAI to make it easier to do government paperwork.

What it does

LaporAI is a simulation of an Indonesia tax file system, Coretax, using fictional taxpayer data and synthetic documents.

The website exposes its filing functions as state-aware WebMCP tools. An agent can:

• Upload tax file related documents.
• Find missing or inconsistent assets and liabilities.
• Navigate through the filing steps.
• Prepare evidence-based corrections for the taxpayer to approve or reject.
• Validate the return and explain what still needs attention.

The final reviewer remains human-only. The agent operates the website's functions while the taxpayer sees and controls the result.

How we built it

I built LaporAI with React, TypeScript, and Vite. The application runs entirely in the browser with no backend or database.

I studied the Indonesia's Coretax system and recreated the parts needed for an end-to-end demonstration. The filing workspace includes documents, employment income, assets, liabilities, family and PTKP (Personal tax-free allowance) information, validation, declaration, and a simulated receipt.

Challenges we ran into

The current WebMCP input model accepts structured JSON, but it does not provide a native File or Blob input for PDF attachments.

To support agent-assisted document import, I created an upload_tax_document tool that accepts the document type, file name, and Base64-encoded PDF bytes. LaporAI reconstructs the file in the browser, validates it, extracts the supported records, and checks that it belongs to the active fictional taxpayer.

Accomplishments that we're proud of

I created a complete end-to-end tax-filling workflow with human in the loop to make sure the data is valid.

What we learned

WebMCP enables the agent to easily move through the web just like a normal human. This enables a lot more UX possibilities.

What's next for LaporAI

The next step is to test the same approach on other public-service workflows where people face complex forms and must verify sensitive information, including visa and immigration applications, passport applications, business registration, healthcare reimbursement, and student financial aid.

The bigger goal is to show public institutions that adding agent-readable capabilities does not require removing human control. A government service can let an agent handle the paperwork while the citizen remains the final decision-maker.

Built With

• codex
• react
• typescript
• vite