AI Assistant Concept

AI Assistant Concept
Photo by Mojahid Mottakin / Unsplash

Artificial Intelligence and all the tools built upon it are extremely hyped. Implementing production-ready applications at this phase of AI development can be difficult.

The main problem is to have a stable application, which will be user-friendly. However, as developers, we can build tools for ourselves pretty easily.

We must connect programming skills with AI APIs or frameworks, which are responsible for developing applications powered by language models.

One of these frameworks is LangChain, described in the next paragraphs.

Introduction | 🦜️🔗 Langchain
LangChain is a framework for developing applications powered by language models. It enables applications that:

AI Assistant Structure

How should it look like?

Before implementation of an AI Assistant, it is necessary to ask yourself questions:

  • for what do I need it,
  • What are my programming skills?
  • Maybe I should use no code tools?
  • How much time can I put into building it?
  • What tools, and apps do I use every day, maybe they have API interfaces so that I can connect them with AI.

Below I will present you the current implementation of my AI Assistant.

schema of AI assistant technologies

The above diagram will make more sense when I present the implemented features.

Let's start with tracking home budgets and expenses.

Tracking Expenses

Tracking expenses is a headache, personally, I tried to sit down with Excel and scan my receipts or banking history. Then categorize that and count. It requires a lot of work, but...

Last weekend I found the solution which will be glad to share.

Capturing expenses

According to the diagram of the flow, let's go deeper into the steps:

  • make a photo of a receipt (automation proceeds in every language, but the example is in polish. However, the output is always translated to English)
example receipt in Poland
  • start shortcut
shortcut of the adding expense

The above shortcut can be set to be run on voice or a combination of buttons.

It extracts the text from a photo, which looks like this:
F.H.U. BOSMAN NEJSCIE NA PLAZE NR 1 84-120 NEADYSLAMOWO (B) NIP 5871543717 nr: 4035 PARAGON FISKALNY WODA 1szt. ×5,00 5,00A FUZETEA 1szt. x6,00 6,00A PINO 0,5L. 1szt. x8,00 8,00A SOK DO PIWA * 1szt. ×1,00 1,00B DORSZ FIL.SMAZ.ZST 1szt. x27,00 27,00B FLADRA TUSZ. SMAZ. ZESTAL 1szt. ×25,00 25,00B SANDACZ FILET SMAZ. ZESTAH 1szt. ×31,00 31,00B SPRZEDAZ OPODATKONANA A 19,00 SPRZEDAZ OPODATKONANA B 84.00 1 , PTU A 23% 3,55 PTU B 8% 6,22 SUMA PTU 9,77 SUMA PLN 103,00 ROZLICZENIE PLATNOSCI GOTOAKA 103,00 PLN 00073 #1 KELNER 2021-07-04 14:36 539BE78CAA52CE22792DBE9896233813385284F6

This text is passed to the API as a body of a request.

  • API

This script proceeds the text into the correct output, which is Posted to Notion Budget Tracker.

  • Notion Budget Tracker
notion updated expenses

As can be seen, expenses are added one by one with one of the categories that were defined.

The presented code is still in MVP, but the above idea will be wrapped in a product and launched in the upcoming weeks.

The current implementation also assigns the category of an expense that is fetched directly from Notion, to avoid assigning some random categories which would be difficult to track in charts, etc.

Notion Budget Tracker

Get it now

📔 Book of the week

Last week, many of the creators decided to read the book "Million Dollar Weekend" by Noah Kagan.

Still waiting for my delivery to share with you about the content, but below you can find an interesting video with the review:

🐦 Tweet of the week

Simon Høiberg (@simonhoiberg) on Threads
Be careful about work without working. → Redesigning your website → Improving your workflows → Trying out new technology → Reading about business → Doing “research” → Planning Not saying these can’t be important. But ask yourself honestly... Are they, right now? Often enough, they’re work without working. They make you feel like you’re making progress, but in reality, you’re getting no where.

NotionFlow

Get Demo

📽️ Last Video

🚀 Thank you for reading, till the next time!