Curio
An autonomous AI teaching assistant that listens, understands, and generates educational aids in real-time during lectures. It captures audio input, extracts semantic structures, and dynamically serves visual maps, quizzes, and summaries to students.
Launch Specifications
Product Overview
Curio empowers teachers by automating the creation of educational content during live lectures. It listens to lecture speech, references pre-loaded textbook context, extracts key pedagogical concepts, and generates structured learning materials on the fly.
- Real-time text & visual content generation.
- Adaptive lecture context understanding.
- Multi-modal classroom outputs (slides, mind maps).
- Instant test & quiz generation based on spoken points.
What Curio Can Generate
Visual concept maps from lecture context.
Auto-generated assessment questions.
Concise key-point digests.
Interactive cards for review.
Structure slides matching lecture flow.
Printable worksheet exercises.
The Problem
Teachers spend an average of 10-15 hours a week manually drafting quizzes, summaries, and lesson materials, taking away crucial time from student mentorship. Existing static tools do not adapt to live class discussions.
Our Solution
Curio listens to live lectures, maps spoken words to a dense educational database using hybrid semantic search, and dynamically builds high-fidelity educational outputs, reducing prep time by 80%.
Technical Architecture
Curio operates as a multi-agent system built on LangGraph. An ingestion node handles speech-to-text streams, passing chunks to an analyzer agent. The analyzer performs semantic search in PostgreSQL (pgvector) to grab relevant reference context, and routes structural requests to secondary generation nodes.
Ingestion & Processing Pipeline
Tech Stack
Dashboard View Simulation
Your AI Co-Teacher, in Every Lesson.
Listening to: "Machine Learning Basics - Decision Trees and Entropy..."
Visual concept map linking Entropy with Information Gain.
1. What is entropy used to measure in decision trees?
- Information gain splits data
- Leaf nodes hold outputs
- Pruning prevents overfit
Key Engineering Challenges
- •Mitigating hallucination of historical and mathematical facts in front of students.
- •Handling real-time web socket streaming latency over fluctuating school Wi-Fi networks.
- •Ensuring clean speaker diarization in a noisy classroom setting.
Key Lessons Learned
- ✓Rigid evaluation sets (Ragas) are critical before updating active model prompts.
- ✓Caching repetitive retrieval vectors at the local Redis layer saves up to 35% in API costs.
- ✓Simple UI indicator overlays build student trust during brief text-generation delays.
Development Roadmap
Real-time Auditory Pipeline
WebSocket audio translation layers.
Multi-Agent Refinement
Splitting slide creation and quiz creation agents.
LMS Integrations
Syncing materials to Google Classroom and Canvas.
Related Products
Every product is built with a focus on solving real problems.
Interested in engineering collaboration, specialized quantitative models, or custom educational AI solutions? Let's connect.