Savant
Safe, contextual AI for classrooms.
Empowering teachers to lead AI experiences.
In mid-2025, I co-founded Savant, an AI platform designed to give educators control over how artificial intelligence showed up in their classrooms.
Insight
AI had entered the classroom. Teachers hadn't been invited.
The proliferation of general-purpose AI tools created an immediate tension in education: students had access to powerful models, and institutions had no framework for managing them. The dominant response from educators was restriction — banning tools, flagging usage, treating AI as a threat to learning integrity rather than an opportunity to extend it.
The problem wasn't the tools. It was the interaction model. Student and AI, with the teacher entirely absent. No visibility. No guardrails. No context. A generic model has no knowledge of your syllabus, your pedagogy, or your students. It will answer an exam question as readily as a study question, and hallucinate a concept as confidently as it explains one.
I believed the paradigm could be inverted: rather than keeping AI out of the classroom, give educators the tools to bring it in on their own terms. Savant's core proposition was a new relationship to AI — student, teacher, and AI all working together to improve the learning experience.
Idea
Put the teacher in control.
Customer research shaped a product built around four connected problems:
Misinformation. Generic models carry no classroom context. They sometimes produce confident, plausible, wrong answers — answers that instructors then have to spend time undoing. Savant addressed this by grounding each AI instance in custom knowledge uploaded by the teacher: syllabi, lecture notes, reading lists, and primary sources. The model knew what it was supposed to know, and nothing more.
Invisibility. Educators had no window into how AI was being used by their students. Savant gave teachers a dashboard of student interactions — frequently asked topics, engagement patterns, and the ability to view individual conversations — turning AI usage from a blind spot into a source of insight about where students were struggling.
No guardrails. Standard AI models have no concept of academic integrity. Savant introduced structural guardrails: the model would not directly answer questions it identified as belonging to assignments, quizzes, or exams (uploaded to its knowledge base by teachers). Instead, it redirected students toward concepts and approaches — supporting learning rather than replacing it. We also prototyped cheating detection by cross-referencing assignment submissions against student prompt histories.
Safety was built into the foundation. Savant flagged concerning content — mental health signals, inappropriate usage — for teacher and administrator review, and enabled threaded student-teacher-AI conversations so educators could enter a dialogue at any point.
My co-founder led market research, business development interviews, and the financial and business case analysis. I owned product vision, customer use cases, deeper UX discovery interviews, and built the working product — including the LLM integrations and core platform architecture.
Impact
A sound product. An honest reckoning.
We built a fully functional prototype, conducted substantive customer research with educators, and validated genuine demand for the problem Savant was solving. The feedback on the product direction was strong.
The economics told a different story.
Our market sizing and unit economics analysis — conducted using both top-down and bottom-up models across operating budgets and teacher-equivalent salary frameworks — pointed to a serviceable addressable market in the range of $82–186M for US private high schools. A real market, but not large enough for venture-scale.
Compounding that: the cost of operating custom AI systems per institution was structurally higher than what schools could bear. Sales cycles and pilot periods in education are long. The unit economics didn't close, and the gap wasn't a product problem — it was a market reality.
Raising venture capital to bridge these gaps would have required fund return profiles the market sizing couldn't support.
We made the decision to shut down Savant 6 months in. It was neither a failure of product thinking nor of execution. It was the application of the same analytical rigor that shaped the product that led to the decision to shut the business down.
