README.md

metadata

title: Astral.AI
emoji: 🏃
colorFrom: pink
colorTo: red
sdk: gradio
sdk_version: 5.37.0
app_file: app.py
pinned: false
license: afl-3.0
short_description: An AI Powered Tracker

🩺 Pediatric Respiratory Triage Assistant (NLP-Powered)

An intelligent, interactive assistant that helps parents and caregivers triage common pediatric respiratory symptoms.
Built using interpretable machine learning, custom NLP pipelines, and deployed with a friendly Gradio chatbot interface.

⚠️ This tool provides non-diagnostic guidance only. It is not a substitute for medical advice.

---

🚀 Live Demo

Try the app on Hugging Face Spaces
(replace with actual URL when deployed)

---

🎯 Project Objective

• Enable users to describe symptoms in natural language
• Use an ML model to classify input into one of four triage levels
• Return a safe, easy-to-understand recommendation
• Support early triage decisions, especially in low-resource or high-volume contexts

---

🧠 How It Works

1. User enters a free-text symptom description
2. The text is processed via TF-IDF vectorization
3. A Decision Tree classifier predicts the triage category
4. A friendly chatbot message is displayed based on the prediction

Triage Labels:

Label	Guidance
🟢 Monitor at Home	Mild symptoms, low risk. Watch and observe.
🟡 Consult GP	Suggests seeing a doctor for further evaluation.
🫁 Use Inhaler	Asthma-like symptoms. Use prescribed inhaler and monitor closely.
🔴 Visit Emergency	Serious symptoms. Seek urgent medical attention.

---

🧪 Model Summary

• Vectorizer: TfidfVectorizer with bigrams
• Best model: DecisionTreeClassifier (Accuracy: 96%)
• Other tested models: Linear SVM, XGBoost, LightGBM, Random Forest

---

🗃️ Files in This Repo

├── app.py # Gradio app entry point
├── best_model_decision_tree.joblib # Trained ML model
├── requirements.txt # All required libraries
├── README.md # You're reading it

---

🧰 Tech Stack

• Gradio – chatbot interface
• Scikit-learn – model training + TF-IDF pipeline
• XGBoost, LightGBM, Random Forest – tested alternates
• Joblib – model persistence
• Matplotlib, Seaborn – EDA & diagnostics

---

🧠 Limitations

• Not a diagnostic system — designed only for low-risk triage advice
• Based on synthetic and heuristic-labeled data from medical education text
• Requires expansion for multilingual or multi-condition support

---

🧭 Future Plans

• Integrate BERT for deeper semantic understanding
• Add class weighting or active learning
• Train on verified clinical text or real-world triage logs
• Publish as an embeddable widget or API

---

⚠️ Disclaimer

This project is for educational and research purposes only. It is not a certified medical device or diagnostic tool.
Always consult a licensed healthcare provider for professional medical advice.

---

📄 License

MIT License – feel free to modify, fork, and improve responsibly.

---

🙌 Credits

Built with ❤️ by SilverDragon9
Inspired by pediatricians, caregivers, and the need for accessible, responsible healthcare AI.