bunqSplit + MILA

• 

  The architecture for the LangGraph

• 

  bunqSplit: Hassle free item-by-item split

Inspiration

We targeted a wider audience to automate their lifestyle using bunq to avoid small hassle and also promote a better lifestyle.

What it does

bunqSplit uses the SOTA OCR model by MistralAI to detect items in a receipt item-by-item where you can easily split the money amongst friends using the latest bunqMe link which is a killer for Tikkie. We allow to reach multiple users without the need for users signing up on bunq. Furthermore, with the use of MILA that runs on a multi-agentic framework in LangGraph, we make end-to-end automation by accessing the user transactions for the week and then sending them notifications by classifiying their spending into different categories by the bunq, then we proceed to do the find a pattern in their spending and also find the outliers, we follow this by using a researcher by Tavily API and finally the recommendation agent provides a notification on the bunq app.

Challenges we ran into

The biggest challenges were the integration of the LangGraph framework in Swift. Another problem that we ran into was the NVIDIA API for the LLMs due to the high variation in the model responses.

Accomplishments that we're proud of

We are proud to integrate the bunqAPI with production ready code by integrating multiple features in the app that is directly targeted to a wide user group. Furthermore, we are proud of using LangChain framework to build the multi-agent that is end-end for user transactions. We are also exceptionally proud to have a working app in SwiftUI which is propreitary in Apple with limited documentations.

What we learned

We learnt that there is a big scope for bunq autonomous landscape by adding nudges for the users for a better lifestyle. We also see the potential for this app to be integrated in the services like Tricount.

What's next for bunqSplit + MILA

We are thinking to integrate the MCP server for the LangGraph and add LangGraph to the SwiftUI.

Github

link

Built With

• agent
• ai
• api
• flask
• langchain
• langgraph
• mistral
• nvidia
• ocr
• python
• rest
• swift