TLDR: JELAI is an open-source platform that integrates AI tutoring with learning analytics directly within Jupyter Notebooks. It captures detailed student interactions (code, errors, chat) to provide context-aware AI support, enabling personalized feedback and facilitating research into student-AI interactions and pedagogical strategies.
In the evolving landscape of education, the integration of Artificial Intelligence (AI) holds immense promise, particularly for providing personalized student support. However, many existing AI tools often fall short in understanding the specific learning context of a student, leading to generic responses that may not truly aid the learning process. This challenge is particularly evident in programming education, where students often use environments like Jupyter Notebooks.
Addressing this gap, researchers have introduced JELAI (Jupyter Environment for Learning Analytics and AI), an innovative open-source platform designed to seamlessly blend fine-grained Learning Analytics (LA) with Large Language Model (LLM)-based tutoring directly within the familiar Jupyter Notebook environment. JELAI aims to provide context-aware AI scaffolding and facilitate in-depth research into how students interact with these intelligent tools.
What is JELAI and Why is it Important?
JELAI is built on a modular, containerized architecture. It features extensions for JupyterLab that handle telemetry (detailed tracking of student actions like code edits, executions, and errors) and chat interactions. A central middleware component processes this learning analytics data and enriches LLM prompts with real-time context. This means the AI tutor, named Juno within JELAI, can offer more relevant and timely assistance by understanding what a student is currently working on, their recent mistakes, and their conversation history.
The platform’s significance lies in its ability to capture integrated code interaction and chat data. This rich dataset is crucial for both providing immediate, context-sensitive AI support and for researchers to study student behavior patterns, help-seeking strategies, and the effectiveness of different AI configurations. Unlike other systems that might operate separately from the main coding environment, JELAI integrates directly, offering a holistic view of the student’s learning journey.
How JELAI Works: A Glimpse into its Architecture
JELAI’s system comprises four main components working in harmony:
- User Notebook Container: This is where the student’s JupyterLab instance resides, equipped with extensions to capture detailed interactions and provide the AI tutor interface.
- Middleware Container: The brain of JELAI, this component processes all incoming student activity and chat data. It also prepares context-rich prompts for the LLM, applying pedagogical rules defined by instructors.
- JupyterHub Container: Manages user access, creates individual learning environments for students, and ensures scalability for multiple users.
- LLM Server: This component interfaces with the chosen language model. JELAI supports local open-source LLMs like Llama3.1 and Gemma2 via Ollama, prioritizing data privacy, but can also connect to external APIs.
When a student interacts with their notebook or the AI tutor, JELAI logs these actions. The AI tutor then receives messages enriched with immediate context, allowing the LLM to generate responses that are highly relevant to the student’s current task and learning state. This ensures that the AI’s feedback is not just helpful, but also pedagogically sound.
Demonstrating JELAI’s Capabilities
Preliminary evaluations have showcased JELAI’s feasibility and potential. In terms of performance, JELAI demonstrated stable operation with acceptable response times for local LLMs, and its JupyterHub deployment successfully supported over 20 concurrent users with low resource consumption, indicating good scalability.
Two proof-of-concept use cases further illustrated its power:
- Help-Seeking in a Python Course: JELAI was deployed in an introductory Python course, successfully logging student code interactions and chat messages with the LLM tutor. This allowed researchers to categorize help-seeking behaviors (e.g., instrumental vs. executive help) and explore relationships between coding actions and help-seeking patterns.
- Prompt Comparison Pilot: This study demonstrated JELAI’s utility for A/B testing different AI configurations. Novice programmers used JELAI with either a generic “helpful assistant” prompt or a “pedagogical” prompt. The results showed that the pedagogical prompt led to longer dialogues and more instrumental help requests, suggesting a more engaged and productive learning process.
These demonstrations confirm JELAI’s ability to provide real-time responsiveness, identify actionable help-seeking signals, and enable rapid experimentation with AI designs without requiring significant infrastructure changes.
Also Read:
- Integrating MOOCs, Smart Teaching, and AI for a Unified Educational Experience
- Unraveling Complex Student Errors: A New Approach to Diagnosing Combined Steps in Tutoring Systems
Looking Ahead
JELAI represents a significant contribution as an open-source technical framework for integrating learning analytics and AI tutoring within Jupyter Notebooks. While powerful local LLMs can be computationally intensive, ongoing advancements in models are mitigating this. Future work for JELAI includes enhancing AI interaction with multi-step processing, simplifying configuration, integrating advanced techniques like Retrieval-Augmented Generation (RAG) and AI Agents, and improving interoperability with educational standards. The ultimate goal is to further investigate how JELAI can foster effective help-seeking and improve learning outcomes.
For more detailed information, you can refer to the full research paper available here.
