Overview

This repository houses a comprehensive approach to developing an AI assistant tailored for K12 STEM education, with a strong emphasis on culturally responsive teaching practices. Our project spans several interconnected components: the generation of a culturally responsive dataset, prompt engineering for an AI assistant, and the implementation of a Retrieval-Augmented Generation (RAG) model using PostgreSQL and pgvector.

Repository Structure

1. Dataset Preparation

   • Focuses on generating a culturally responsive dataset for fine-tuning a Large Language Model (LLM).
   • Techniques: PDF text extraction, segmentation, and Q&A generation.
   • Tools: UnstructuredPDFLoader, OllamaFunctions LLM model.
   • Output: JSON file with Q&A pairs that reflect diverse voices and inclusive practices.

2. Prompt Engineering

   • Details the creation and refinement of prompts to develop an AI assistant that supports culturally relevant K12 STEM education.
   • Techniques: System prompt design, iterative feedback, and testing.
   • Tools: Llama 3.1 model on the Ollama platform.
   • Focus: Inclusivity, engagement, cultural relevance, and continuous improvement.

3. RAG Implementation

   • Implements a Retrieval-Augmented Generation (RAG) model that utilizes PostgreSQL and pgvector for vector embedding storage.
   • Techniques: Document chunking, vector embedding generation, cosine similarity retrieval.
   • Tools: PostgreSQL with pgvector, LLaMA-3 model.
   • Output: Contextually accurate and relevant responses to user queries based on retrieved text chunks.

1. Objective and Scope

Objective: To develop an AI assistant and tools that help educators create culturally relevant, engaging, and inclusive lesson plans for K12 STEM education.

Scope: - Design and test initial prompts for culturally responsive AI. - Generate a culturally responsive dataset for fine-tuning. - Implement and refine a RAG model for efficient information retrieval and augmentation.

2. Detailed Components

2.1 Dataset Preparation

• Objective: Generate a dataset that reflects diverse voices and inclusive practices in STEM education.
• Techniques:
  • Text Extraction: Extract meaningful text segments from PDFs using UnstructuredPDFLoader.
  • Text Segmentation: Apply fixed-size and sliding window segmentation to maintain context and structure.
  • Q&A Generation: Use OllamaFunctions to create Q&A pairs focused on culturally responsive teaching.
  • Data Cleaning and Export: Structure the dataset into JSON, with options to convert to CSV for easier integration.

2.2 Prompt Engineering

• Objective: Design prompts that guide the AI assistant to generate culturally relevant lesson plans.
• Approach:
  • System Prompt: Emphasizes inclusivity, engagement, and cultural relevance.
  • Testing and Feedback: Continuous refinement based on feedback to improve cultural nuance handling and engagement.
• Enhancements:
  • Updated prompts for clarity and specificity.
  • Revised structures for broader cultural relevance.
  • Ongoing evaluation to identify and close gaps.

2.3 RAG Model Implementation

• Objective: Implement a RAG model using PostgreSQL and pgvector for efficient retrieval of contextually relevant information.
• Techniques:
  • Vector Embedding Generation: Extract and store embeddings in PostgreSQL with pgvector.
  • Cosine Similarity Retrieval: Retrieve the most relevant text chunks based on user queries.
  • Augmentation: Use retrieved chunks to provide contextually accurate responses with the LLaMA-3 model.

3. Future Plans

1. Continuous Feedback Loop: Engage with educators and experts for ongoing review and improvement.
2. Fine-Tuning: Create and refine a dataset for further fine-tuning based on identified gaps.
3. Expanded Testing: Continue testing and documenting results to close remaining gaps.