Updated to reflect recent changes

README.md

@@ -6,129 +6,140 @@ colorTo: yellow
 sdk: gradio
 sdk_version: 5.44.1
 app_file: app.py
-pinned: false
+pinned: true
 python_version: '3.10'
 short_description: Advanced prompt engineering for educational AI systems.
 thumbnail: >-
   https://cdn-uploads.huggingface.co/production/uploads/68700e7552b74a1dcbb2a87e/Z7P8DJ57rc5P1ozA5gwp3.png
 ---
 
-# EduBot: Educational AI Assistant
+# Mimir: Educational AI Assistant
 ## Advanced Prompt Engineering Portfolio Project
 
 ### Project Overview
-EduBot demonstrates sophisticated prompt engineering techniques applied to educational technology, showcasing the implementation of context-aware AI systems that prioritize pedagogical effectiveness over simple answer generation. This project exemplifies professional-grade prompt design for educational applications with built-in academic integrity safeguards.
+Mimir demonstrates sophisticated prompt engineering techniques applied to educational technology, showcasing the implementation of context-aware AI systems that prioritize pedagogical effectiveness over simple answer generation. A key feature is its ability to **dynamically generate custom data visualizations**, determined by an intelligent decision engine that assesses whether a visual aid will enhance the pedagogical explanation. This project exemplifies professional-grade prompt design for educational applications, embodying the role of an educational partner that guides students to discover answers for themselves.
 
-### Technical Architecture
+***
 
+### Technical Architecture
 **Core Technologies:**
-- **LangChain**: Prompt template management and conversation chain orchestration
-- **Gradio**: Full-stack web interface with custom CSS styling
-- **Hugging Face Inference API**: Model deployment and response generation
-- **Python**: Backend logic and integration layer
+
+* **LangChain**: Prompt template management and conversation chain orchestration.
+* **LangGraph**: Orchestrates the application's flow as a state machine (**StateGraph**). It manages the conditional logic for the tool-use decision engine, routing user queries between the LLM, a pre-built **ToolNode** for graph generation, and the final response node.
+* **Gradio**: Full-stack web interface with custom CSS styling.
+* **Hugging Face Inference API**: Model deployment and response generation.
+* **Python**: Backend logic and integration layer.
+* **Matplotlib**: Powers the dynamic, in-memory generation of educational graphs and charts.
 
 **Key Frameworks:**
-- `langchain.prompts.ChatPromptTemplate` for dynamic prompt construction
-- `langchain_huggingface.HuggingFaceEndpoint` for model interface
-- `langchain.schema` message objects (HumanMessage, AIMessage, SystemMessage)
+
+* `langchain.prompts.ChatPromptTemplate` for dynamic prompt construction.
+* `langchain_huggingface.HuggingFaceEndpoint` for model interface.
+* `langchain.schema` message objects (HumanMessage, AIMessage, SystemMessage).
+* `langgraph.graph.StateGraph` & `langgraph.prebuilt.ToolNode` for building and executing the conditional logic graph.
+* `langgraph.checkpoint.memory.MemorySaver` for persistent conversation state.
+
+***
 
 ### Prompt Engineering Techniques Demonstrated
+#### 1. Unified System Prompt Architecture
+Employs a single, comprehensive system prompt that establishes the AI's core persona as **Mimir, an expert multi-concept tutor**. This foundational prompt meticulously defines the AI's behavior, tone, and pedagogical mission. It integrates:
+
+* **Core Educational Principles**: A directive to prioritize teaching methodology, foster critical thinking, and provide comprehensive explanations over direct answers.
+* **Defined Persona & Tone**: Specific instructions to maintain an engaging, supportive, and intellectually appropriate tone for high school students, while avoiding fluff and emojis.
+* **Specific Response Guidelines**: Contextual rules for handling different academic tasks, such as explaining concepts in math problems instead of solving them, or discussing research strategies for essays rather than writing them.
+
+#### 2. Instructional Design Integration
+The core prompt incorporates evidence-based instructional design principles:
 
-#### 1. **Context-Aware Template Selection**
-Implemented intelligent subject detection algorithm using keyword analysis to dynamically select appropriate prompt templates:
-```python
-def detect_subject(message):
-    # Keyword-based classification system
-    # Routes to specialized educational templates
-```
-
-#### 2. **Role-Based System Prompting**
-Four distinct prompt templates employing specific pedagogical roles:
-
-- **Mathematics Tutor Template**: Emphasizes conceptual breakdown and process explanation
-- **Research Skills Mentor Template**: Focuses on source evaluation and methodology guidance  
-- **Study Skills Coach Template**: Incorporates learning style optimization and retention strategies
-- **General Educational Assistant Template**: Comprehensive academic support framework
-
-#### 3. **Instructional Design Integration**
-Each template incorporates evidence-based instructional design principles:
-- **Scaffolding**: Breaking complex concepts into manageable components
-- **Socratic Method**: Guiding discovery rather than providing direct answers
-- **Metacognitive Strategies**: Teaching learning-how-to-learn approaches
-
-#### 4. **Academic Integrity Constraints**
-Implemented ethical AI guidelines through prompt engineering:
-- Explicit instructions to avoid homework completion
-- Focus on process over product delivery
-- Critical thinking skill development emphasis
+* **Scaffolding**: Breaking complex concepts into manageable components.
+* **Socratic Method**: Guiding discovery rather than providing direct answers.
+* **Metacognitive Strategies**: Teaching learning-how-to-learn approaches.
+
+#### 3. Academic Integrity Constraints
+Implemented ethical AI guidelines directly into the system prompt:
+
+* Explicit instructions to avoid homework completion.
+* Focus on **process over product delivery**.
+* Critical thinking skill development emphasis.
+
+#### 4. Two-Stage Tool-Use Prompting
+A sophisticated two-stage prompting strategy governs the use of the `Create_Graph_Tool`:
+
+* **Tool-Use Decision Prompt**: A highly-constrained template is used by the `Tool_Decision_Engine` to determine if a tool should be used. This prompt forces a **YES** or **NO** response based on whether a visual aid would significantly enhance learning, using explicit **INCLUDE** and **EXCLUDE** criteria.
+* **Tool-Execution Guidance**: The main system prompt contains separate, explicit instructions on how to use the tool once the decision has been made. It provides the exact **JSON structure** the model must output, including fields like `data`, `plot_type`, and `educational_context`, ensuring the generated graphs are pedagogically sound.
+
+***
 
 ### Advanced Implementation Features
+#### Intelligent Graphing Tool Integration
+A custom, dynamic visualization system was developed to provide multi-modal educational responses.
 
-#### **Conversation Memory Management**
-```python
-# LangChain message history integration
-messages = [SystemMessage, HumanMessage, AIMessage]
-# Maintains educational context across interactions
-```
-
-#### **Response Streaming & Truncation**
-- Smart text truncation preserving sentence integrity
-- Real-time response streaming for improved UX
-- Error handling and fallback mechanisms
-
-#### **Template Chaining Architecture**
-```python
-chain = template | llm
-response = chain.invoke({
-    "question": message,
-    "system_message": educational_context
-})
-```
+* **LLM-Powered Analysis**: For relevant queries, a targeted LLM call is made using the specialized YES/NO decision prompt.
+* **Dynamic Visualization Tool (`Create_Graph_Tool`)**: Designed and implemented a custom visualization tool using **matplotlib**. The tool receives a JSON configuration from the LLM and generates high-quality bar, line, or pie charts. The entire process occurs in-memory:
+    * The plot is rendered into a `BytesIO` buffer.
+    * The image is encoded into a **base64 string**.
+    * The final output is an HTML `<img>` tag with the embedded base64 data, which is displayed directly in the chat interface, eliminating the need for file I/O.
+    * The tool's docstring provides a clear schema and usage instructions for the LLM, ensuring reliable and pedagogically sound visualizations.
 
-### User Interface Engineering
+#### Stateful Conversation Management with LangGraph
+Implements persistent, multi-turn conversations using LangGraph's **MemorySaver**. This allows the application's state, including the full message history (`add_messages`), to be saved and resumed, ensuring robust context management even when tool use is involved.
 
-#### **CSS Grid System Implementation**
-- Viewport-based height allocation (15% title, 60% chat, 25% input)
-- Full-width responsive design
-- Cross-browser compatibility optimization
+#### Response Streaming & Truncation
+* Smart text truncation preserving sentence integrity.
+* Real-time response streaming for improved UX.
+* Error handling and fallback mechanisms.
 
-#### **Component Architecture**
-- Modular Gradio component structure
-- Custom CSS class integration
-- Accessibility-compliant design patterns
+#### Template Chaining Architecture
+The core logic utilizes **LangChain Expression Language (LCEL)** to pipe inputs through templates, models, and tools.
+
+***
+
+### User Interface Engineering
+* **Gradio Layout & Custom Styling**: The interface is built with `gr.Blocks`, using `gr.Column` and `gr.Row` to structure the main components. A custom `styles.css` file is loaded to apply specific theming, responsive design, and layout rules, moving beyond default Gradio styling for a tailored user experience.
+* **Component Architecture**: Modular Gradio component structure with custom CSS class integration and accessibility-compliant patterns.
+
+***
 
 ### Prompt Engineering Methodologies Applied
+* **Template Parameterization**: Dynamic variable injection for contextual responses.
+* **Persona-Driven Response Generation**: Crafting a detailed persona within the system prompt to guide the AI's tone, style, and pedagogical approach consistently.
+* **Domain-Specific Language Modeling**: Educational vocabulary and pedagogical terminology integration.
+* **Multi-Modal Response Formatting**: Structured output generation with educational formatting.
+* **Agentic Tool Routing**: Designing prompts and logic that enable an AI system to intelligently decide which tool is appropriate for a given task, simulating agent-like behavior.
 
-1. **Template Parameterization**: Dynamic variable injection for contextual responses
-2. **Behavioral Constraint Definition**: Explicit instruction sets for ethical AI behavior
-3. **Domain-Specific Language Modeling**: Educational vocabulary and pedagogical terminology integration
-4. **Multi-Modal Response Formatting**: Structured output generation with educational formatting
+***
 
 ### Professional Applications
-
 This project demonstrates competency in:
-- **Enterprise-Grade Prompt Design**: Scalable template architecture
-- **Educational Technology Integration**: Pedagogically-informed AI system design
-- **Ethical AI Implementation**: Academic integrity safeguards and responsible AI practices
-- **Full-Stack AI Application Development**: End-to-end system implementation
 
-### Technical Specifications
+* **Enterprise-Grade Prompt Design**: Scalable template and tool-use architecture.
+* **Educational Technology Integration**: Designing AI tutors with robust pedagogical frameworks and dynamic, multi-modal response capabilities.
+* **Ethical AI Implementation**: Academic integrity safeguards and responsible AI practices.
+* **Full-Stack AI Application Development**: End-to-end system implementation.
+* **Intelligent Agent & Tool Development**: Building AI agents that can utilize custom tools to solve complex problems.
 
+***
+
+### Technical Specifications
 **Dependencies:**
-- `langchain` - Prompt orchestration and conversation management
-- `langchain-huggingface` - Model integration layer
-- `gradio` - Web application framework
-- `huggingface_hub` - Model deployment interface
+
+* **Core ML/AI**: `transformers`, `torch`, `accelerate`
+* **LangChain & LangGraph**: `langgraph`, `langchain-core`, `langchain-community`, `langchain-huggingface`
+* **UI Framework**: `gradio`
+* **Visualization**: `matplotlib`, `plotly`, `pandas`, `numpy`, `scipy`
+* **Utilities**: `python-dotenv`
+* **Monitoring**: `langsmith` (Optional)
 
 **Deployment:**
-- Hugging Face Spaces compatible
-- Environment variable configuration for API keys
-- Production-ready error handling and logging
 
-### Results & Impact
+* Hugging Face Spaces compatible.
+* Environment variable configuration for API keys.
+* Production-ready error handling and logging.
 
-EduBot represents a synthesis of prompt engineering best practices with educational technology requirements, demonstrating the ability to create AI systems that enhance rather than replace human learning processes. The project showcases advanced technical implementation while maintaining focus on pedagogical effectiveness and academic integrity.
+***
 
----
+### Results & Impact
+Mimir represents a synthesis of prompt engineering best practices with educational technology requirements. The integration of an intelligent, conditional graphing tool demonstrates the ability to create AI systems that augment and enhance human learning processes, embodying the role of an educational partner who empowers students to succeed through genuine understanding.
 
-**Portfolio Demonstration**: This project evidences advanced prompt engineering capabilities, full-stack AI application development, and domain-specific AI system design suitable for enterprise educational technology environments.
+> **Portfolio Demonstration**: This project evidences advanced prompt engineering capabilities, full-stack AI application development, and domain-specific AI system design suitable for enterprise educational technology environments.