QoutiOussama13 / VentureBuddy

An AI-powered advisor for everyone interested in Entrepreneurship dedicated for the Moroccan public

Mokawil .AI 🤖💼🇲🇦

Problem

Starting and managing a business in Morocco is challenging for entrepreneurs and aspiring business owners due to a lack of access to reliable information. Many struggle with the complexities of setting up a company, understanding tax requirements, and grasping Morocco-specific business terminology. Business failures in Morcco skyrocketed in 2022, rising by 17.4% compared to 2021.

Intorducing Mokawil.AI

Mokawil .AI is the first-of-its-kind GenAI-powered chatbot designed to assist entrepreneurs and aspiring entrepreneurs in starting their businesses in Morocco. It has access to specific information covering everything from creating a company, running it, understanding when and how much to pay taxes, business terminology in Morocco, and much more.

Approach:

We utilized the Retrieval-Augmented Generation (RAG) technique to enhance the accuracy and reliability of our generative AI models. RAG optimizes the output of a large language model by referencing an authoritative knowledge base outside of its training data sources before generating a response.

The main tool we employed for making the RAG is Langchain 🦜️⛓️ for performence and simplicity .

Phase 1: Data Collection and Embedding

1. Data Collection:

   • Gathered detailed blogs on entrepreneurship,taxes,and everything related to starting a company from BHAdviser.

2. Embedding:

   • Utilized a model from Hugging Face to transform data into vector representation.

3. Vector Storage:

   • Stored embeddings in the vector database FAISS.

Phase 2: User Interaction

1. User Interaction:

   • Users input prompts through the Streamlit interface.

2. Prompt Embedding:

   • Embedded user prompts using the same model used for data embedding.

3. Similarity Search:

   • Compared embedded user prompts with vector database data to find the best match.

Phase 3: Large Language Model (LLM)

1. LLM Selection:

   • Utilized the Mistral-7B-OpenOrca language model, incorporating Q4 quantization.

2. Answers Generation:

   • Fed results of similarity search to the LLM to generate answers based on user-specific needs and available information the answer are also saved in chat memroy to make use of them later .

3. User Interface (UI):

   • Presented answers in the Streamlit UI for user consumption.

Global Architechtures

Instructions

To run the Mokawil .AI application, follow these steps:

1. Open Google Colab:

   • Access Google Colab using your preferred web browser.

2. Install Requirements:

   • Inside the Colab environment, install the necessary requirements. You can do this by running the following command in a code cell:

     !pip install -r requirements.txt

3. Run the Notebook:

   • Open the notebook named "Rag financial advisor GPU.ipynb" in the Colab environment.
   • Execute the cells in the notebook to run the application.

Alternatively, you can explore a demo of the application directly on Hugging Face : Geek Team - Hugging Face Note : it takes time for generation for low ressouce

Demo

Future work

we plan on making Mokawil.AI more advanced and more helapful for the users by making it :

• Personalized Responses:

  • Provied tailord responses based on user-specific information.

• Multilingual Support:

  • to be Capable of understanding and responding in multiple languages.

• Notification System:

  • Sends notifications for important deadlines to keep users informed.

• Versatile Input Support:

  • Accepts text, audio, and document inputs for flexibility.

• Financial Document Analysis:

  • Analyzes financial documents to provide insightful information.

• And Much More...

Geek Team

• Oussama Qouti
• Mahmoud Bidry