T&T & Body Builder

Build your body without thinking. A weekly food purchase plan with discounted prices which is generated by AI from T&T flyers.

Comment

Inspiration

Why T&T? Because it's the closest supermarket to my house. I've spent a lot of money at T&T, and now I need an internship. I hope T&T notices me.

I'm broke, so I don’t want to spend too much money on food.
I'm skinny, and I want to gain muscle, so I need to know how many calories I’m eating.
I'm lazy, too lazy to think about what to eat. So I usually just buy chicken breast. I get tired of it, but I’m still too lazy to think of alternatives.

So, I’m creating a tool to solve all of the above.

What it does

It gives you suggestions based on T&T’s flyer. You enter your weekly calorie goal, and AI will generate a shopping list for you. Since it’s based on the flyer, most of the items are on sale.

How I built it

  1. I used Flask + its built-in Jinja templating for the frontend/backend. I haven’t really gone deep into web development, but I’ve built websites with AI before, so this part was pretty easy for me.

  2. I used Airtop as my main web scraping tool — it has built-in LLM.
    Playwright is another tool I used, mainly to cache rendered structures.

  3. OpenAI SDK and Gemini 2.5 Pro API.

Challenges I ran into

Right now the AI tends to overestimate calorie counts. In the future, I’ll create another AI agent to serve as an evaluator.

Accomplishments that I'm proud of

This is the first website I’ve built that uses AI agents as its core feature.
Even if no one else uses it, I will.

What I learned

Structured output from AI agents is the thing I learned that excites me the most.
If we control it well, we can build a lot of useful products with it.

Also, I’ve learned a lot about the limitations of AI (especially LLMs). In some parts of my code, I still use traditional scraping techniques (like manually analyzing the HTML structure and fetching specific parts).
I didn’t do this because AI can't do it, but because — for this MVP — doing it manually was faster and saved me time exploring the AI side too deeply in vain perhaps.

What's next for T&T & Body Builder

  1. Add a custom prompt field so the AI can consider user-specific requests.
  2. Add discounted products links to the product names in the result page.
  3. Try to improve the accuracy of calorie estimation.
  4. If T&T is willing to offer me an API for sale items, that would be perfect.
  5. Add a feature for users to input their height, weight, and target weight.
  6. Arrange my code properly.
  7. Make my AI agents more modular — like using tools and workflow.
  8. Deploy it and keep updating this until it becomes a real product.

Built With

  • ai-agent
  • airtop
  • flask
  • gemini-api
  • openai-sdk
  • playwright
  • python
  • web-scraping
Share this project:

Updates