Work
Built on a decade of shipped work
Tools and research already reaching 15 million students across the US and India. Everything we build produces open data, open tools, and open research — evidence for the whole field.
Open Items
Open assessment infrastructure for K-12 education
The open applied layer on top of open education infrastructure. 34K+ CC-licensed assessment items with AI generation, LLM evaluation, adaptive practice, and 18 interactive math widgets — built on the CZI Learning Commons Knowledge Graph. The first serious applied project demonstrating what's possible when you build openly on shared infrastructure.
- —Built on the CZI Learning Commons Knowledge Graph (250K standards, 2K learning components, 273K relationships)
- —LLM-as-Judge evaluation pipeline: 5-dimension scoring, 85% auto-approve rate, 98% mathematical accuracy
- —9 item types: multiple choice, multiple select, T/F, short answer, numeric, essay, cloze, matching, ordering
- —AI content generation using Gemini 3 Flash — full K-12 curriculum at $25-50 total cost
- —Adaptive practice with Elo-rated difficulty calibration
- —PDF worksheet generation with interactive widget renderers
- —REST API and embeddable SDKs for integration by other tools
SmartPaper
Bridging paper and digital learning at national scale
Computer vision tool that lets teachers print worksheets, students write by hand, and AI scores instantly. Deployed across government schools in Rajasthan, India — bridging the gap between paper-based classrooms and digital assessment infrastructure.
- —Deployed in partnership with Indian state governments in Rajasthan
- —Measurable reduction in learning poverty across participating schools
- —UNESCO-recognized assessment innovation
- —Generates open psychometric data on assessment items at scale
PlayPower
Free games, quests, and AI tools reaching 10M+ students
50+ free K-8 math games, 45 high school quests, AI teacher tools — distributed by Savvas Learning to schools nationwide. The commercial track record that demonstrated the need for open infrastructure.
- —Distributed by Savvas Learning (one of the largest US K-12 publishers)
- —Free games at playpowergames.com covering half of K-8 math topics, English and Spanish
- —Teacher AI tools: lesson planning, activity planning, worksheet creation, text leveling
- —Experience building at scale informed the design of Open Items
UpGrade
Open-source A/B testing for education
An open-source experimentation platform purpose-built for educational software, enabling researchers and developers to run rigorous A/B tests within learning environments. Originally built on Carnegie Learning's infrastructure.
- —Funded by the Gates Foundation and Schmidt Futures
- —Built on Carnegie Learning's infrastructure
- —Purpose-built for educational experimentation
- —Supports multi-armed bandit and factorial designs
AI Research Tools
Qualitative and quantitative instruments for studying AI in education
A suite of research tools for studying AI's impact — including AI-powered qualitative interviews (text and voice), adaptive testing platforms, and synthetic student simulation for rapid psychometric feedback.
- —AI-powered qualitative interviews by text and voice
- —Adaptive testing engine with real-time item selection
- —Synthetic student simulation for instant psychometric calibration
- —Open evaluation frameworks for assessing AI tool quality
Selected publications
75+ publications across top venues
LLM Difficulty Estimation
200 experimental conditions, 15+ large language models evaluated for math item difficulty estimation.
A/B Testing at Scale Workshops
Four annual workshops convening researchers on experimentation methodology in learning at scale.
CHI Honorable Mention
Large-scale experiment with 70,000+ subjects on design choices in educational games.
Full publication list spanning CHI, AIED, EDM, L@S, JCAL, and more. Venues include ACM, Springer, IEEE, and ISLS.
See where we're headed
Our roadmap for the next phase of research and tools.