AYA

Overview

AYA was an EdTech startup leveraging reinforcement learning-optimized NLP to extract and formulate questions from academic video content. Selected for NYU Entrepreneurial Institute Summer Sprint (Top 7% - 10 of 150 startups).

The Problem

Students struggle to efficiently learn from video lectures:

• Passive consumption: Videos don’t encourage active recall
• Note-taking burden: Manual extraction is time-consuming
• Question generation: Creating practice questions requires expertise

Technical Approach

Speech Processing Pipeline

• Automatic speech recognition for lecture transcription
• Speaker diarization for multi-speaker content
• Timestamp alignment for navigation

NLP for Question Generation

• Reinforcement learning optimization for question quality
• Entity and concept extraction from transcripts
• Difficulty calibration based on content complexity

Educational Design

• Bloom’s taxonomy-aligned question types
• Spaced repetition integration
• Progress tracking and analytics

Achievements

Technical Stack

Speech:     Whisper, pyannote
NLP:        Transformers, RL-based generation
Backend:    Python, FastAPI
Frontend:   React

Lessons Learned

• Product-market fit: Educational tools require deep integration with existing workflows
• User research: Teachers and students have different needs
• Startup execution: Speed of iteration matters more than initial feature set

Impact on Career

The experience building AYA informed my approach to production AI systems at Mount Sinai—understanding that technical capability must align with user needs and deployment constraints. The speech processing and NLP skills directly transferred to clinical voice agent development.