VanTax MX

Image of chat response when Receipt is sent.
Main dashboard of the user.
Landing page of the project.

Inspiration

We were enticed by the difficult path that a new person on the job market has to follow on Mexico when it needs to deduce taxes, or even make an anual declaration. Rules change over time, so even taking a course on this topic can lead to legacy knowledge that needs to be relearned

What it does

The user takes a picture of their ticket or invoice, the assistant makes a plan on what to do next, even creating reports to the DB so that they can be accessed in the monthly reports

How we built it

Tech stack

MongoDB Atlas
Gemini API
NestJS
NextJS

Challenges we ran into

How to make an LLM agnostic agent?
What rules should we consider to feed the LLM?
Who limit the public to focus the solution?

Accomplishments that we're proud of

Connecting the LLM via the NestJS API endpoint so the user can interact with it
Learning how to make the LLM output based on our needs
Establish the interaction workflow between components: FrontEnd and Chatbot with BackEnd and LLM
Catalog tickets or invoices as Monthly or Annual declaration using LLM

What we learned

How an AI agent works
How can it interact with its environment
The strict rules to follow to make a fiscal movement
How focus our idea in specific problem to attend it using new technologies

What's next for VanTax MX

Adding other AI models to the platform so the user can pick up the most they want, even bringing their own API keys to reduce token costs on our side
Automating completely the process of annual declaration
Handle most complicated situations about fiscal process