SAGE (Sustainable Advisory & Guidance Expert) is an AI-powered chatbot providing intelligent insights on sustainability. It offers two chatbot versions:
-
DistilBERT-Based Chatbot: Uses intent recognition and NLP for structured responses.
-
GPT-Powered Chatbot: Utilizes OpenAI’s GPT-3.5 API for dynamic, context-aware answers.
This project compares accuracy, response quality, and effectiveness in aiding sustainability decision-making.
🌍 Whether you're exploring eco-friendly solutions, carbon footprints, or sustainable lifestyle choices, SAGE has you covered!
Ensure you have the necessary dependencies installed:
If you're using Conda, create or update your environment:
conda env create -f requirements.yml # For a new environment
conda env update --file requirements.yml --prune # Update existing environment If using pip, install all dependencies:
pip install -r requirements.txtTo use the GPT-powered chatbot, get an OpenAI API key:
🔗 Sign up at OpenAI API
🛠 Set your API key in the terminal:
export OPENAI_API_KEY="your-api-key-here"make all✅ Train DistilBERT Chatbot (NLP Model)
make train✅ Run DistilBERT Chatbot (Intent-Based)
make chat✅ Run GPT-Powered Chatbot
make gptuiI did 10 epochs, but you can try doing them from 5 to 7. to reduce time.
| Epoch | Loss | Accuracy | F1-score |
|---|---|---|---|
| 1/10 | 3.1377 | 29.83% | 29.03% |
| 2/10 | 1.2033 | 72.03% | 68.80% |
| 3/10 | 0.9780 | 88.68% | 88.09% |
| 4/10 | 0.2234 | 96.07% | 96.02% |
| 5/10 | 0.1583 | 97.68% | 97.54% |
| 6/10 | 0.0810 | 98.14% | 98.08% |
| 7/10 | 0.2626 | 98.69% | 98.65% |
| 8/10 | 0.0526 | 98.77% | 98.78% |
| 9/10 | 0.2280 | 99.03% | 99.01% |
| 10/10 | 0.0319 | 98.86% | 98.84% |
🔹 Training Screenshot:
📌 Rapid learning: Accuracy jumped from 29.8% to 88.6% in 3 epochs
📌 Stabilized performance: After 6 epochs, accuracy reached 98%+
📌 Final accuracy: ~99%, indicating optimal chatbot readiness
📌 Overfitting prevention: No additional training required
🔹 Uses a pre-trained NLP model
🔹 Matches user input to predefined intents
🔹 Returns structured, rule-based answers
🔹 Works offline after training
🔹 Uses OpenAI’s GPT model
🔹 Generates real-time, contextual responses
🔹 Handles open-ended, complex questions
🔹 Requires an active internet connection & API key
| Feature | DistilBERT Chatbot | GPT-3.5 Chatbot |
|---|---|---|
| Response Type | Rule-based | Contextual & Dynamic |
| Accuracy | High for structured queries | Adaptive & evolving |
| Training Needed? | Yes (pre-trained NLP model) | No (API-based AI) |
| Best Use Case | Quick predefined answers | Deep, AI-driven explanations |
By comparing the two models, SAGE explores the advantages of rule-based AI vs. deep learning-based conversational AI in addressing sustainability queries.
🖥️ Graphical Interface:
- A user-friendly chatbot window
💬 Interacting with SAGE:
- Type a question & hit enter → Your query disappears, and SAGE provides an AI-generated response.
- Click "Exit" → The chatbot closes gracefully.
🤖 GPT Mode Features:
- Handles complex and open-ended queries better than the DistilBERT model.
- Generates long-form, contextual answers based on AI training data.
- Uses real-time GPT interaction instead of predefined responses.
- 🌍 Sustainability-Focused: AI-powered insights on eco-friendly solutions.
- 🤖 Machine Learning at Work: Combines intent recognition, NLP, and GPT AI for better understanding.
- 🎯 User-Centric Design: A clean, intuitive chatbot experience with two AI models for comparison.
SAGE is more than just a chatbot—it’s a demonstration of AI’s evolving role in sustainability education and decision-making. 🌱💡
🔹 Ready to explore AI-driven sustainability? Run SAGE today! 🚀