AI POWERED MENTOR MENTEE MACHING SYSTEM

Overview

This project aims to develop an AI-powered system for alloting the right mentors to students based on sharesd interests, alignment of goals, mentor availability. It leverages a encoded vector based similaity searching technique. BERT's advanced encoding technique helps make the vectors contextually rich and appropriate.

---

Implementation Details

1. Mock Dataset Generation:

   • Generated mock dataset of 300 mentors with 6 key attributes.

     • Expertise
     • Weaknesses
     • Interests
     • Teaching Style
     • Professional Goals
     • Availability

   • All the attibutes are a key indicator of a mentor profile. With carefully crafted prompts to GPT-4o model, a pseudo-realistic dataset was crafted.

   • Combining this dataset with student dataset from the AI-powered-personalised-learning-path-for-students task, we get two rich datasets containing explicit information about students and mentors.

2. Feature Engineering:

   • Explicit feature engineering is done on both student and mentor datasets.
   • This adds more information about an object (read: students and mentors) and help make the vectors more information rich.

3. Vector DB powered similarity search:

   • Google's BERT model is used to encode the textual database into vector collections with the schema

   {"id" : <vector_representation>}
   

   • The solution involves the use of MilvusDB to store these vectors. MilvusDB provides in-built similarity search techniques that can be used to find the perfect mentor for each student.

---

How to Run?

1. Clone the Repository:

   git clone https://github.com/coder-utkarshchaudhary/AI-powered-mentor-mentee-matching-system.git 
   cd AI-powered-mentor-mentee-matching-system

2. Install Dependencies: Ensure pip is installed, then run:

   pip install -r requirements.txt

3. Set Environment Variables: Create a .env file in the project root and add your OpenAI API key:

   OPENAI_API_KEY=your_openai_api_key
   

   NOTE: This is only needed if you want to make a new dataset from scratch.

4. Run the project: In the main.ipynb file, press Run All. NOTE:
   1. Ensure that you have a GPU to run BERT tokenizer.
   2. BERT tokenizer takes some tme to download, so allow the notebook to run completely.

---

Final Submission

• Codebase: Link to GitHub repository
• Technical Document: Link to project report

---

Notes

1. This project uses GPT-4o model. The model's API is access via On-Demand platform. The code will not directly work for OpenAI API calls directly. Please modify the get_llm_response function in datagen.py.

2. BERT needs a GPU for faster inference. On each run, the BERT configurations download from huggingface. Be patient while the model is downloading.

   Future Scope:

   1. A RAG system can be incorporated on mentor and student datasets for Agent powered mentor-mentee matching.
   2. Implementation of ML models like Linear Regression, XGBoost, ANNs etc. for assigning custom weights to custom embeddings generated using BERT.
   3. Implementation of a UI system with user input importance for search criteria.

---

Acknowledgement

Thanks to the open-source libraries and tools used in this project --> OpenAI, MilvusDB, Torch, Huggingface and On-Demand.
This project was made as a part of the Alcovia Intern Optional Task - 1. I extend a heartfelt thank you to the Alcovia team for providing me this opportunity.

---

License

This project is licensed under the MIT License. You are free to use, modify, and distribute this software, provided proper attribution is given.

Prerquisite project

Also check out this project's first part at AI-powered-personalised-learning-path-for-students