Professor Starstuff 🚀✨

🌌 Bringing Astronomy to Life for Kids

Professor Starstuff is a multimodal AI chatbot that makes learning about space fun and interactive for children. This chatbot leverages natural language processing, vector-based retrieval, and podcast-style responses to engage young minds with fascinating space facts.

---

🔥 Features

• 🧠 Natural Language Processing (NLP): Understands and responds to kids' astronomy questions.
• 📚 Vector-based Knowledge System: Retrieves accurate space facts from YouTube video transcripts.
• 📡 NASA Image API: Fetches real images of celestial objects for better visualization.
• 🎙️ Podcast-Style Responses: Generates engaging storytelling audio from text-based answers.
• 🗂️ ChromaDB Integration: Efficient search and retrieval of astronomy knowledge.
• 🔊 OpenAI TTS: Converts text responses into audio format.
• 🚀 Deployabled on Heroku: Django-based backend with an HTML/CSS/JavaScript frontend.

---

🏗️ Tech Stack

Backend:

• 🟢 Django - Main backend framework
• 🔵 SQLite (ChromaDB) - Vector database for storing astronomy facts
• 🔴 Redis - Cloud memory storage for conversation context
• 🟣 Heroku - Deployment platform

AI & Retrieval:

• 🤖 GPT-4 & GPT-3.5 Turbo - Language models for chatbot responses
• 📌 ChromaDB - Vector storage for RAG (Retrieval-Augmented Generation)
• 📡 NASA API - Fetches real space images
• 🔊 OpenAI TTS - Text-to-speech for podcast-style responses

Frontend:

• 🌐 HTML, CSS, JavaScript - Simple, interactive UI
• 🎨 Bootstrap - Styling framework

---

📊 Dataset & Processing

Professor Starstuff is built on a dataset extracted from YouTube astronomy video transcripts:

1. Transcript Extraction: Uses youtube_transcript_api to fetch video transcripts (~8 hours of content).
2. Chunking Strategy:
   • Chunk size: 500 tokens
   • Overlap: 100 tokens for better context retention
3. Vector Embeddings:
   • Uses text-embedding-3-large from OpenAI for high-quality embeddings.
4. Storage:
   • Stored in ChromaDB with metadata (e.g., video titles) for efficient retrieval.

---

📡 System Architecture

1. User Input: Professor Starstuff processes questions and determines if they are related to astronomy.
2. Decision Making (GPT-4):
   • If the question is astronomical, it proceeds to retrieval.
   • If general, it provides a direct response.
3. Retrieval & Response Generation:
   • ChromaDB fetches relevant facts.
   • NASA Image API retrieves space-related images.
   • OpenAI TTS converts responses into audio.
4. Final Output:
   • Provides a text response, space images, and an audio podcast snippet.

---

🚀 Deployment

Django + ChromaDB on Heroku

1. Clone the repository:

   git clone https://github.com/Senimtra/astronomy-bot.git
   cd astronomy-bot

2. Install dependencies:

   pip install -r requirements.txt

3. Run the application locally:

   python manage.py runserver

4. Deploy to Heroku:

   heroku create professor-starstuff
   git push heroku main

---

📈 Evaluation & Optimization

Professor Starstuff is continuously evaluated using LangSmith:

• ⚡ Inference Time: Measures response speed.
• 📚 Retrieval Efficiency: Ensures accurate fact retrieval.
• 🔧 Tool Efficiency: API calls (NASA, ChromaDB, OpenAI TTS).
• 📊 Model Selection:
  • GPT-4: Best for decision-making.
  • GPT-3.5 Turbo: Faster for general responses.

---

🌟 Future Improvements

• 📡 Live Space Event Integration: Fetch real-time astronomy news.
• 🔊 Voice Interaction: Enable full voice-based conversation.
• 🛠 Streaming Responses: Faster and smoother podcast delivery.
• 🎓 Educational Quizzes: Make learning more interactive.
• 👤 User Profiles: Personalize experience based on learning history.

---

🎉 Thanks for Exploring with Professor Starstuff!

Made with 💙 for young space explorers! 🌠