Career Mentor Chatbot

This project demonstrates a Career Mentor Chatbot that leverages a fine-tuned large language model (Mistral-7B) to provide career advice and answer career-related queries. We fine-tuned a gated base model (Mistral-7B) using LoRA (Low-Rank Adaptation) on a curated career Q&A dataset, and built a Flask backend to serve the model. Our React-based frontend is still under development.

Check out the model files here on Hugging Face:

https://huggingface.co/sp4rsh/career_mentor_finetuned_new/tree/main

Overview

The goal of this project is to create a useful chatbot that helps users with questions such as:

• How to switch careers (e.g., from marketing to data science)
• How to start a career in AI research
• What companies hire Robotics Engineers
• Salary insights for various roles (e.g., Salesforce Administrator)

We achieved this by:

• Fine-tuning a large pre-trained model using LoRA for parameter-efficient adaptation.
• Testing the model in a Jupyter Notebook (see Career_Mentor_V2.ipynb) with sample prompts.
• Building a Flask backend (app.py) that loads the model and exposes a /generate API endpoint.
• Starting development on a React frontend to provide a user-friendly chat interface.

Fine-Tuning Process

1. Dataset Preparation:
   We curated a dataset of career Q&A pairs. Each entry was formatted as:

   • Prompt:
     Question: <your question here>\nAnswer:
   • Completion:
     The corresponding answer.

2. Model Adaptation using LoRA:
   We applied LoRA for parameter-efficient fine-tuning on the base model (Mistral-7B). This process allows us to adapt the model with minimal additional storage, as only the adapter weights are saved.

3. Testing in Jupyter Notebook:
   In Career_Mentor_V2.ipynb, we loaded the tokenizer and the fine-tuned model, then tested it using several prompts:

   prompts = [
       "Question: How do I switch from marketing to data science?\nAnswer:",
       "Question: How can I begin working as an AI Researcher?\nAnswer:",
       "Question: What kind of companies hire Robotics Engineers?\nAnswer:",
       "Question: What is the average salary for a Salesforce Administrator?\nAnswer:"
   ]
   
   for prompt in prompts:
       inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
       outputs = model.generate(**inputs, max_new_tokens=100)
       response = tokenizer.decode(outputs[0], skip_special_tokens=True)
   
       print(f"Prompt:\n{prompt}")
       print(f"Response:\n{response}")
       print("="*50)

Backend

• The Flask backend (app.py) is responsible for: • Loading the Model:

• It loads the base model in 4-bit quantization and applies the LoRA adapter. • Providing an API Endpoint:

• A POST endpoint /generate accepts a JSON payload with a career question and returns a generated response. • CORS Support: -The backend uses flask-cors to allow cross-origin requests from the React frontend.

• Example usage (via cURL):

curl -X POST -H "Content-Type: application/json" \
     -d '{"prompt": "How do I switch from marketing to data science?"}' \
     http://localhost:5000/generate

Frontend

• The React frontend is under development and will be available soon.