Lumix AI

AI Agent for Tutors: Smarter Grading, Planning, and Student Insights

Comment

🌟 Lumix AI — Your AI Teaching Agent

AI-Powered Tutor Assistant — Teaching brilliance, powered by an intelligent AWS Agent.

Lumix helps tutors manage students, grade worksheets, generate personalized lesson plans, and organize their teaching workflow — all through an AI agent built with AWS Bedrock and Textract.

💡 Inspiration

As a tutor, I spend hours grading, planning, and managing student progress — time that could be spent actually teaching. I wanted to build an AI agent that acts as a teaching co-pilot. When planning most of my features, I would always think back on what I found annoying (i.e. creating worksheets, looking through past papers) and find a way to automate them.

⚙️ What it does

Lumix is an AI-powered tutoring assistant that supports the entire teaching workflow:

  • Interactive Agent Chat: Tutors can ask: “Plan next week’s algebra lesson for Alice based on her last worksheet.” The AI agent orchestrates data from student profiles and the question bank to produce a plan.
  • Automated Grading: Uses AWS Textract for OCR and Bedrock for answer understanding and feedback generation.
  • Performance Insights: Tracks student progress and recommends follow-up topics dynamically.
  • Question Bank Management: Upload PDFs or past papers → AI extracts, classifies, and tags questions by topic/difficulty.

🏗️ How we built it

Lumix is built as a monorepo with two core services:

Web App Service — lumix-web

  • Framework: Next.js 15 (App Router) + TypeScript + Basic CRUD Services
  • Styling: Tailwind CSS
  • Database: DynamoDB
  • Storage: S3
  • Integration: Calls AI service endpoints

AI Service — lumix-ai

  • Framework: FastAPI (Python 3.11)
  • AI Models: AWS Bedrock (Amazon Nova, Claude, Titan)
  • OCR: AWS Textract for document parsing
  • Deployment: AWS Lambda via Mangum
  • Containerization: Docker

🚧 Challenges we ran into

  • Bedrock Agent orchestration: Balancing custom logic (FastAPI) with Bedrock’s built-in reasoning and data access was tricky.
  • Document parsing accuracy: OCR layouts from worksheets varied a lot — we had to normalize Textract outputs for consistent grading.
  • Prompt consistency: Generating structured, curriculum-aligned lesson plans required multi-prompt chaining and schema validation.
  • Lambda deployment: Packaging FastAPI with dependencies like boto3 and mangum for Lambda required custom Docker layers.

🏆 Accomplishments that we're proud of

  • Built a fully functional AI agent that can plan lessons, grade papers, and analyze student progress.
  • Integrated AWS Bedrock + Textract + DynamoDB into a coherent workflow.
  • Created a clean Next.js 15 interface for tutors to interact with the AI agent seamlessly.

📚 What we learned

  • How to build multi-service architectures on AWS and the power of Bedrock Agents in chaining multiple AI tools and APIs into a single reasoning flow.
  • How to design prompts and data pipelines for reliable educational AI applications.
  • The importance of grounding AI outputs with contextual data from student performance and history.

🚀 What’s next for Lumix AI

  • Student Portal: Give students AI-powered feedback and personalized practice sets.
  • Curriculum Integrations: Connect to real syllabi (IGCSE, IB, AP) for tailored content.
  • Mobile App: On-the-go grading and lesson planning for tutors.
  • Plugin Ecosystem: Extend to schools and tutoring centers via secure APIs.

Built With

Share this project:

Updates