This is a submission for the Heroku "Back to School" AI Challenge
What I Built
I built Teacher Assistant, a web application that helps educators generate structured lesson plans in minutes. Teachers often spend significant time every day designing lesson plans, which can be repetitive and time-consuming. This application leverages AI to streamline the process: educators simply upload the document they intend to teach (e.g., a textbook chapter, reading material, or notes), and the system generates a context-aware lesson plan tailored to the grade level, topic, and duration specified by the teacher.
The generated lesson plan includes:
- Step-by-step teaching activities
- Required materials
- Timed sections for better pacing
- Formative quiz questions with answers
- Differentiation for two levels (support & extension)
This helps teachers save time and focus more on meaningful student interaction rather than paperwork.
Category
Educator Empowerment
Demo
Checkout the project here: Teacher Assistant
Flow of the application:
- Teacher uploads a PDF document of the lesson content.
- The system extracts text and creates vector embeddings, stored in a pgvector-powered database.
- Teacher specifies topic, grade and duration.
- With one click, a tailored lesson plan is generated, aligned with the uploaded material.
Screenshots
Screenshots showcasing the Teacher Assistant UI, document upload workflow, and generated lesson plans.
How I Used Heroku AI
The application integrates multiple Heroku AI features:
1. Heroku Managed Inference and Agents
- Embeddings Generation: Used Cohere embeddings API to convert uploaded document text into embeddings.
- Lesson Plan Generation: Used Claude Sonnet 4 to generate structured lesson plans based on both teacher input (grade, topic, duration) and retrieved contextual embeddings.
2. pgvector for Heroku Postgres
- All embeddings are stored and queried in a pgvector-enabled Postgres database.
- This ensures semantic search capabilities to retrieve the most relevant content from the uploaded material during lesson plan generation.
Together, these components enable a seamless multi-agent workflow: one agent handles document embedding and storage, while another focuses on contextual lesson plan generation.
Technical Implementation
- Frontend: React + TypeScript, with Tailwind CSS for styling.
- Backend: Express + TypeScript
-
AI Workflow:
- On document upload, text is extracted and embeddings are generated via Managed Inference (Cohere).
- Embeddings are stored in Postgres with pgvector extension for efficient semantic retrieval.
- When the teacher requests a lesson plan, embeddings are queried to extract the most relevant context, and an agent (Claude Sonnet 4) generates the structured plan.
-
Multi-Agent Architecture:
- Document Agent: Handles embedding creation and storage.
- Planning Agent: Generates the lesson plan using retrieved context and teacher inputs.
Database: Heroku Postgres with pgvector.
Challenges Solved
Efficient context retrieval: Implemented semantic search over lesson documents using pgvector to ensure lesson plans are accurate and aligned with actual teaching material.
Multi-agent coordination: Designed a workflow where one agent enriches the knowledge base (embeddings) and another produces actionable lesson plans.
User experience: Created a minimal, intuitive interface where teachers can generate professional-grade lesson plans with a single click.
Results
I have pitched and demoed Teacher Assistant to few educators. The response was overwhelmingly positive, several teachers expressed strong interest in adopting it immediately, finding it a valuable tool to save time and improve teaching quality. The application demonstrates how AI can meaningfully empower educators while improving classroom experiences.
Top comments (9)
Hey everyone! If anyone tried uploading a document and it failed, there was a temporary issue with the object storage service. It’s now been fixed. Please let me know if you face any further issues.
A very splendid work.
Thank you.
This is great, thank you so much for sharing your experience!
Thank you! Glad you liked it.
I think you are the winner for teacher category for sure.
That means a lot coming from you. I’m just excited to see how it all turns out.😅
what do you think of mine:
dev.to/prime_299792/study-mate-ai-...
be honest, can I really win in student or creative category? 😅
Thanks for sharing! I’ll definitely check it out. I’m sure your project has a lot of great aspects, but honestly, it’s all up to the judges. Best of luck with it!