Dalelk: The Official AI Academic Assistant for the College of Computer Science and Engineering for Jeddah University

 
├── classifier_training/               # The brain that understands what the user is asking about
│   ├── certifications/                # Questions certifications folder
│   ├── courses/                       # Questions courses folder
│   ├── general/                       # Questions general folder
│   ├── classifier_dataset.xlsx        # Used to train the intent classifier
│   ├── create_dataset.py              # A script that automatically generates dataset questions
│   └── Evaluation_100.xlsx            # Dataset for training/evaluation
│
├── datasets/
│   ├── Certifications.xlsx            # Dataset about professional certifications
│   ├── Courses_new.xlsx               # Dataset for the new course plan
│   ├── QA.xlsx                        # General academic Q&A dataset
│   └── Students Survey.xlsx           # Dataset built from students questions
│
├── frontend/                          # Frontend website implementation with responsive UI and API integration
│   ├── public/
│   ├── src/                           # Main source code (components, pages, hooks)
│   ├── index.html
│   ├── package.json
│   ├── package-lock.json
│   ├── postcss.config.js
│   ├── tailwind.config.ts
│   ├── tsconfig.app.json
│   ├── tsconfig.json
│   ├── tsconfig.node.json
│   └── vite.config.ts
│
├── models/                            # Trained AI models used by Dalelk
│   └── fine_tuned_marbert/            # Fine-tuned, trained on Dalelk's datasets
│
├── unit testing/                      # Backend API endpoint tests and rate limiting validation
│   ├── main.py
│   ├── unit_test.py
│   └── unit_test_ratelimit.py
│
├── build_embeddings.py                # bulid embeddings for dataset
├── data_loader.py                     # loading and preparing all the datasets into the program
├── llm_inference.py                   # communicating with the LLM to generate the final answer for the user.
├── load_embeddings.py                 # loading the pre-built embeddings
├── logger.py                          # Loading everything that happens in the program
├── main.py                            # the entry point of Dalelk where everything starts and comes together.
├── NOTICE.txt                         # google policy
├── query_classifer.py                 # The query classifier that analyze the user's intent
├── query_classifier_training.ipynb    # Jupyter Notebook where the MARBert model was trained and fine-tuned.
├── requirements.txt                   # lists all the Python libraries that Dalelk needs to run.
├── retrieve.py                        # finding the most relevant information to answer the user's question.
└── setup.py                           # packaging and setting up Dalelk as a Python project.

Getting Started

1. Get Your Google API Key

1. Go to Google AI Studio
2. Sign in with your Google account
3. Click "Get API Key"
4. Click "Create API Key"
5. Choose your project
6. Copy the generated API key
7. Paste it in the project in the .env file in the API_KEY1 field.

2. Get Your Hugging Face API Key

1. Go to Hugging Face
2. Sign in with your account
3. Go to Embedding Gemma and accept the usage policy
4. Click on your profile picture → "Settings"
5. Go to "Access Tokens"
6. Click "New Token"
7. Give it a name and set access to "Read"
8. Click "Create Token"
9. Copy the generated token
10. Paste it in the project in the .env file in the hf_token field.

3. Download the Classifier Model

1. Go to the Releases page of this repository
2. Download the classifier model file
3. Place it inside the models/ folder

4. Extract the Model

1. Extract the downloaded file
2. Make sure the extracted content follows this exact path:

!!! Windows users: !!! When extracting, make sure the final path looks like this and does not have an extra folder: models/fine_tuned_marbert/(model files directly here)