ServeSmart

Competition Scope

Our purpose in this competition to develop an AI project for sustainable cities and green tomorrows.

Requirements

1. Python: To use this code you have to be using Python 3.10 or higher.
2. Dependencies: Install dependencies from the requirements.txt file using the following code:

pip install -r requirements.txt

Running the Application

To get started, follow these steps: Now to run the code, you need to use Streamlit. Run the app.py file in the user interface folder using the following code:

streamlit run app.py

Overview of Code Functionality

Create Account to Sell

Our Streamlit application contains five tabs. The first tab is for creating accounts (for sellers). The instructions to sign up to system:

1. Enter username.
2. Enter password.
3. Enter the IBAN that you will get the payments.
4. Enter your ID no (for security).
5. Enter your business adress (for security).

User Interface:

Sell Product

The second one is for product selling (adding). The instructions to add product:

1. Enter informations about your meal (AI will improve it).
2. Take a photo of your meal.
3. Enter the price of your meal.
4. Enter your IBAN.
5. Enter your ID no (for security).

User Interface:

Search Product

The third tab provides a search product feature. Here’s how it works:

1. Enter the product name you want to buy and search it.
2. Don't forget learn the ID of the meal you will buy.

User Interface:

Buy Product

After decide the meal that you will buy, you're going to buy it from buy product page.

1. Enter the ID of the meal.
2. Enter your adress.
3. Enter your ID no.
4. Enter the CVV of your card.
5. Enter your card number.

User Interface:

See Your Products

As a seller, you can see your products and analysis from this page. Here what should you do:

1. Enter your username.
2. Enter your password.
3. After login, you can see your product and analysis that made for you.

User Interface:

Login	Analysis

Text Generation

We've added a powerful text generation feature in Sell Product tab of our application. At first, we tried to fine-tune Llama and Gemma models, and actually we did it. But there was an option called Gemini. We did the tests and decided to use gemini-1.5-flash. So we decided to use Gemini API. After hitting the "Submit Product" button, we're sending an API request to Gemini (gemini-1.5-flash) and improve the title-description of the meal.

You can see the example.

Result:

Used System Specifications

• 1 T4-15GB GPU
• 12 GB RAM

Warning

You may experience issues running this on less powerful hardware.

Models Used

This project makes use of the following models:

1. gemini-1.5-flash:

   • Source: DeepMind
   • License: API Used - No License Info

2. ahmeterdempmk/FoodLlaMa-LoRA-Based:

   • Source: Hugging Face
   • License: Apache 2.0

3. ahmeterdempmk/Gemma2-2b-E-Commerce-Tuned:

   • Source: Hugging Face
   • License: Apache 2.0

4. ahmeterdempmk/Llama-3.1-8B-Fast-Food-Based-Tuned:

   • Source: Hugging Face
   • License: Apache 2.0

5. Emir Kaan Özdemir - LSTM Based Time Series Model:

   • Source: GitHub
   • License: Apache 2.0

Note

Please ensure compliance with each model's license when using or distributing this project.

Contributors

• Ahmet Erdem Pamuk
• Emir Kaan Özdemir
• İlknur Yaren Karakoç