Lmaghrib-BinYedik

Lmaghrib-BinYedik is a pioneering project developed during the 2024 ThinkAI GenAI Hackathon under the theme of tourism. Our application integrates extensive information about Moroccan monuments into a single, user-friendly chatbot platform, offering tourists a seamless and enriching travel experience.

We Offer:

• Comprehensive Information: Access in-depth descriptions, historical data, and fascinating narratives about Morocco's iconic monuments.
• Navigation Assistance: Utilize precise directions and interactive maps to navigate through Moroccan cities effortlessly.
• Pricing Details: Stay informed with the latest entry fees for monuments, local food prices, and other essential costs.
• User Support: Receive expert travel tips and real-time assistance to make the most out of your journey in Morocco.

Strategy and Execution

Task 1 : Monument Identification

To enhance the accuracy of monument recognition, we evaluated various models before selecting the Google Lens API for its superior performance in handling diverse image data:

Model	Accuracy	Flexibility	Chosen
YoloV8 Classification	High	Low (limited number of classes)	❌
Bing Reverse Search	Average	Average (doesn't work too well with new data)	❌
Google Lens	High	High	✅

Using the Google Lens API, we capture a broad spectrum of images related to each monument. The LLaMA3 LLM then processes this data, efficiently identifying the exact monument name from the image attributes. We selected the 8 billion parameter model of LLaMA3 for its balance of speed and accuracy, crucial for real-time application needs.

Task 2 : Information Retriever

With the monument identified, LLaMA3 adopts a dual role:

Interactive Query Handling: Acts as a responsive agent to field questions about the monument, delivering detailed and accessible information. Data Integration and Management: Leveraging LangGraph, we design complex, stateful interaction chains that maintain conversational context, enhancing user interaction continuity.

Technique	Accuracy	Speed	Relevancy	Chosen
Llama3 Knowledge	High	Fast (doesn't need extra steps to get data)	Low	❌
WebSearch	High	Slow	High	✅

Based on the data retrieved we format it into a suitable format that we can display into the GUI as a markdown with interactive links.

Task 3 : Scam prevention for tourists

Addressing common tourist challenges, we focused on preventing overpayment scams by integrating a reliable pricing guide based on real-time data from scraped local sources like Marjane catalogs.

Technique	Accuracy	Speed	Relevancy	Dependant on Dataset	Chosen
Llama3 Knowledge	Medium	Fast (doesn't need extra steps to get data)	Low	No	❌
Llama3 Prompt Engineered	High(But can only take a small subset of dataset)	Slow (It uses up token for taking the data as context)	High	Yes	❌
WebSearch	High	Slow	High	No	✅
Gemini FineTuned	High	Average	Medium	Yes	✅
CsvAgent	Medium	Fast	Low (Needs Exact Name )	Yes	❌

Additional Features :

Generating map links.

History Persistence: Ensures continuity in user interactions.

Context Relevancy: Keeps conversations relevant and engaging.

On_Chat_Resume: Allows users to pick up where they left off.

Authentication Mechanism: Secures user data and customizes experiences.

Multiple Model Profiles: Offers tailored responses based on user preferences.

Our implementation of LLaMA3 for Lmaghrib-BinYedik presents distinct advantages over cutting-edge models like GPT-4. while GPT-4 offers broad general knowledge, LLaMA3 has been tailored to excel in domain-specific tasks, such as identifying and providing detailed information on Moroccan monuments. This focus ensures that our model not only provides accurate information but does so with a depth of understanding and contextual relevance that is specifically optimized for the tourism sector in Morocco.

Pitch presentation :

Canva Link

Demo :

1337.mp4

Prerequisites

• Python 3.12.0 or newer

Installation

1. Clone the repository:

   git clone https://github.com/Tensor-Titans/Lmaghrib-BinYedik.git
   cd Lmaghrib-BinYedik

2. Create and Activate virtual environments:

   # Create virtual environment
   python -m venv env
   
   # Activate virtual environment (Windows)
   .\env\Scripts\activate
   
   # Activate virtual environment (MacOS/Linux)
   source env/bin/activate

3. Install dependencies:

    pip install -r requirements.txt

4. Set up the environment variables: set the api keys in .env file

   client_id=<your-imagebb-api-key> ---> https://api.imgbb.com/
   hf_token=<your-huggingface-access-token> ---> https://huggingface.co/settings/tokens
   serpApiKey=<your-serp-api-key> ---> https://serpapi.com/manage-api-key
   TAVILY_API_KEY=<your-tavily-api-key> ---> https://app.tavily.com/home
   LITERAL_API_KEY=<your-literal-api-key> ---> https://cloud.getliteral.ai/
   CHAINLIT_AUTH_SECRET=<your-chainlit-secret-key> --->run the command : ( chainlit create-secret )
   at the project root directory to generate the auth secret, and paste it in the.env file
   

Running the Project

1. Start the server:

   chainlit run app.py -w

2. Access the application: Open your browser and go to http://localhost:8000/
   you can use username=admin , password=admin (for the sake of the demo)

Running Tests

Have Fun!

Explore the app, try out all the features, and enjoy discovering all the amazing information about Moroccan monuments, cuisine, and more!

Data Sources

• Moroccan Product Prices Data

• MonumentsMar Dataset

License

This project is licensed under the MIT License. See the LICENSE file for more details.