Run like a
research lab.
Epoch Learn builds the layer beneath AI-native learning: deploying, wrapping, routing, and evaluating small learning models for students, classrooms, cohorts, and researchers. Its engine, Eco Foundry, powers the Free Compute Program that brings that stack to young innovators at no cost.
Why we exist, in three numbers
of Africa’s AI talent has the compute it needs
UNDP, 2024
learning gains from a well-designed AI tutor
Kestin et al., Harvard RCT, 2025
GPU energy from power-capping, for +6.8% time
McDonald et al., MIT Lincoln Lab, 2022
The next epoch of learning runs on
small, accountable models.
Learning is infrastructure
The hard part was never the chatbot. It is the memory, routing, policy, and evaluation around a model: the layer that decides what a learner sees and why. We build that layer.
Hybrid, small, and energy-aware
Light work runs on small on-device models; heavy reasoning routes to our energy-efficient eco servers. Each request goes to the lowest-power option that still meets latency, so capable tutoring reaches learners the cloud prices out.
Structure beats access
A well-designed AI tutor about doubled learning gains in a Harvard RCT (Kestin et al., 2025); a raw, unguided chatbot left students 17% worse off (Bastani et al., 2025). We ship the structure, guardrails, and human review that separate the two.
A restrained system
produces better evidence.
The work is intentionally narrow: a coherent environment, a limited set of studies, and enough operating discipline that evidence stays readable.
Rigor over reach
Smaller, clearer studies until the underlying infrastructure is stable enough to support broader use.
Evidence over theater
No inflated product claims, and no decorative complexity that hides how the system actually behaves.
Learners over traffic
People in the workspace are collaborators in a learning system, not anonymous acquisition metrics.
Small over large
The smallest capable model, run as close to the learner as the task allows, for cost, latency, and privacy.
Start in the open.
Build in private.
Explore the public model playground without an account. Request a pilot workspace to create private cohorts, memory vaults, and guardrailed learning models, or to collaborate on research.