Eedi invests in safety research and tooling to ensure we can be trusted to do no harm. But true safety goes further. It's about demonstrating that learning is happening in measurably safe and effective ways for every student using our products.
Decades of research show that 1:1 tuition is the most effective way to improve learning outcomes, but most students cannot afford it. AI tutoring, supervised by expert human tutors, is Eedi's approach to delivering safety and impact at scale.
AToMs is our proprietary safety system. Designed by Eedi's learning designers it allows us to safely evolve from Human-in-the-Loop (HiTL) to Human-on-the-Loop (HoTL) tutoring, with humans always reviewing what matters most.
AToMS surrounds Eedi Tutor, our constrained AI Tutor, and checks every message as the conversation happens. It reads and reviews messages the AI tutor is about to send, as well as student responses, and decides between a range of next steps: send the message, return it for correction, or escalate it for human review.
Each check carries its own tolerance with safeguarding checks held to the tightest thresholds. Every decision is logged as evidence that AToMS is operating effectively and within agreed limits.
AToMS allows us to employ a human-on-the-loop (HOTL) - approach. Routine exchanges flow straight to the student while messages that carry risk are caught and escalated. At every point, a human tutor is overseeing the system.
AToMS is the difference between promising an AI tutor is safe and being able to demonstrate it, for every message, at any scale, in any language.
