AISTA – AI Teaching & Student Engagement Assistant

AISTA login

Inspiration

The inspiration for AISTA emerged from studying the growing challenges in modern education systems, particularly highlighted in recent reports on the U.S. education sector. Several nationwide observations revealed: Declining academic performance Reading and math proficiency levels are steadily decreasing Over 400,000 classrooms are handled by non-licensed or substitute educators Student disengagement is increasing, reflected in absenteeism, fatigue, and lack of participation Teacher stress and staff shortages are at historic highs Rising mental health and attention-related challenges in classrooms Deep inequities in educational quality, driven by property-tax-based school funding models These issues raised a key question:

How can we ensure consistent, high-quality learning even when teachers are overburdened, unavailable, or replaced by substitutes?

This question led to AISTA — a system designed to keep every learner on track, regardless of classroom constraints.

What It Does

AISTA is an AI-powered Classroom Assistant that supports teachers, substitute instructors, and students through four core capabilities:

🔹 AI Micro-Lectures

Step-by-step explanations of concepts

Real-life analogies and simplified breakdowns

Topic-aligned examples generated on demand

🔹 ClassroomPulse (Engagement Awareness – Prototype Level)

Estimates engagement using visual and interaction-based signals

Designed to observe indicators such as facial orientation, posture, and activity patterns

Implemented as a prototype engagement workflow to demonstrate feasibility

🔹 Adaptive Teaching Support

When engagement drops, AISTA:

Simplifies explanations

Reinforces key points

Introduces short reflective or quiz-style questions

🔹 After-Class Learning Kit

Automatically generated at the end of a session:

Session summary

Structured and simplified notes

Quiz and revision questions

These outputs are generated using LLMs, aligned with how the topic was actually taught.

How We Built It

Building AISTA provided several important technical and design learnings:

** Real-time AI systems are powerful—but complex**

We explored how to connect:

Computer vision pipelines (prototype level)

Large Language Models

Session-based data streams

User interface feedback loops

Measuring engagement is non-trivial

We learned that engagement cannot rely on a single signal. Instead, it benefits from combining multiple indicators, such as:

Face orientation

Expression patterns

Eye focus and head movement

Inactivity bursts and pauses

These insights informed our hybrid engagement design, even where full automation is planned for future work.

Prompt engineering is a critical skill

We refined prompts to:

Avoid repetitive explanations

Adapt examples to lesson context

Generate diverse quiz questions

Produce concise summaries aligned with slide content

User experience matters more than perfect prediction

Even a prototype becomes impactful with:

An intuitive dashboard

Clear, color-coded engagement cues

Simple lesson selection and session flow

How We Built It (Technology Overview)

AISTA integrates LLMs, prototype-level vision logic, and full-stack development.

Tech Stack

Python + FastAPI — Backend logic and API services

Gemini API (LLM) — Micro-lectures, summaries, quizzes

OpenCV / MediaPipe (Prototype / Design-Level) — Engagement signal extraction

Streamlit / Web UI — Live interaction and visualization

NumPy & Pandas — Interaction tracking and pattern analysis

Webcam Input (Optional) — Visual engagement signals

Advanced real-time and vision components are implemented at prototype or design level to demonstrate system feasibility.

Challenges We Ran Into

Engagement values appeared repetitive Early versions produced nearly identical engagement scores. We addressed this using: Time-based smoothing Multi-signal aggregation Repeated-frame filtering LLM responses became repetitive We improved content originality through: Dynamic prompt engineering Context-aware prompt variation Circular generation avoidance Real-time processing latency We optimized performance by: Reducing frame processing frequency Using lightweight vision logic Prioritizing session-level inference Basic dashboard experience We enhanced usability by adding: Low-engagement alerts Concept markers Session timelines Handling multiple lesson topics We introduced: Predefined lesson categories Topic-specific LLM prompt seeds Accomplishments We’re Proud Of We successfully built a working prototype that: Demonstrates AI-assisted teaching support Detects engagement trends at a conceptual level Adapts explanations dynamically Generates learning kits after each session AISTA presents a scalable and ethical AI-assisted education framework that addresses: Teacher shortages Learning gaps Student disengagement