TaxFlow AI

Inspiration

For many taxpayers, simple administrative procedures can become difficult journeys. Registering as a taxpayer or paying an obligation online may require understanding unfamiliar terminology, finding the correct service, preparing documents, and resolving technical errors. At the same time, tax administrations spend significant time answering repetitive questions and processing incomplete requests. TaxFlow AI was inspired by a simple question: can a secure conversational assistant make these procedures easier for citizens while reducing the administrative cost for the State?

What it does

TaxFlow AI is a governed conversational assistant for tax administrations. It focuses on two common procedures:

• taxpayer registration;
• payment of a tax obligation through digital payment.

The assistant identifies the user’s intention, asks only relevant questions, explains the required steps in plain language, and produces a personalized checklist.

When a request cannot be completed automatically, TaxFlow AI converts the conversation into a structured case for a tax officer. The case includes the user’s objective, information already collected, missing elements, and the reason human assistance is required.

The AI provides guidance but does not make binding tax decisions. Sensitive or uncertain cases are explicitly escalated for human review.

How we built it

The project is being built with OpenAI Codex and GPT-5.6.

Codex supports the product and engineering workflow: defining the architecture, implementing the conversational interface, creating the administrative dashboard, testing user journeys, and documenting key decisions.

GPT-5.6 powers the guided conversation. It helps classify user intent, ask contextual follow-up questions, explain procedures clearly, detect missing information, and generate a structured handoff for an authorized officer.

The prototype uses fictional administrative rules and synthetic taxpayer data. It does not connect to a real tax system or process real payments.

Challenges we ran into

The main challenge is balancing simplicity with administrative reliability. A helpful assistant must communicate in accessible language without inventing legal or fiscal rules.

Other challenges include:

• minimizing the collection of personal data;
• preventing the assistant from requesting passwords or banking credentials;
• distinguishing general guidance from an official decision;
• maintaining traceability without exposing sensitive information;
• designing a workflow that remains useful in low-bandwidth environments;
• ensuring that human officers retain validation authority. ## Accomplishments that we're proud of

What we learned

Building TaxFlow AI shows that conversational AI can do more than answer frequently asked questions. It can become the entry point to a complete administrative workflow.

The most important lesson is that responsible public-sector AI requires visible boundaries. Users must know what comes from a configured rule, what was inferred by AI, and what still requires confirmation by a public official.

What's next for TaxFlow AI

The next steps are to:

• complete and test the registration and digital-payment journeys;
• add multilingual and accessibility support;
• introduce stronger security and audit controls;
• evaluate time saved for taxpayers and officers;
• integrate, with authorization, with identity, filing, payment, and case-management systems.

Our long-term goal is to help tax administrations deliver services that are simpler, more accessible, more secure, and less costly to operate.

Built With

• administration
• ai
• api
• automation
• codex
• conversational
• css3
• data
• digital
• fintech
• generative
• government
• govtech
• gpt-5.6
• html5
• human-in-the-loop
• javascript
• language
• natural
• node.js
• openai
• payments
• processing
• tax
• workflow