Three months ago, I connected Evenfield to H.U.N.I.E., our persistent memory system for AI agents. The change fundamentally altered how AI tutoring works in our household.
Before this integration, each tutoring session started from scratch. Claude would assess what my kids knew through questioning, then adapt the lesson. Effective, but inefficient. The AI had to rediscover learning patterns, knowledge gaps, and preferred explanations every single time.
Now the tutor remembers everything. Not just what topics were covered, but how each child learns best. When my youngest struggled with fractions last month, the AI noted her preference for visual representations over abstract explanations. Today, months later, it still opens fraction problems with diagrams before moving to numbers.
This persistent memory transforms the tutoring dynamic. Traditional AI tutoring relies on context windows and conversation history. Useful, but limited. Real tutoring requires understanding that builds over months and years. A human tutor remembers that Sarah grasps concepts through stories while Jake needs concrete examples. They adjust their teaching style accordingly.
H.U.N.I.E. gives Claude this same capability. After every session, the learner agent writes detailed observations to the memory layer. Not just "completed multiplication lesson," but nuanced insights: "tends to rush through word problems without reading carefully," "shows strong pattern recognition in sequences," "gains confidence when encouraged to explain reasoning aloud."
The technical implementation is straightforward. Each tutoring session generates a memory write containing the learner's performance, misconceptions encountered, successful teaching strategies, and emotional state. These observations accumulate in H.U.N.I.E.'s vector database, creating a rich profile that informs every subsequent interaction.
Results show up in subtle but significant ways. The AI no longer suggests review sessions for concepts a child has already mastered. It remembers which explanation style worked for complex topics. When introducing new material, it references previous successes to build confidence.
This matters because homeschool education demands individualization at scale. My three kids learn differently, progress at different rates, and have distinct interests. Traditional curriculum assumes uniform pacing and learning styles. Even adaptive platforms typically reset their understanding of each learner regularly.
Evenfield with H.U.N.I.E. maintains continuity across subjects and time. The same memory system that tracks progress in mathematics informs approaches in science and coding. Cross-subject connections emerge naturally when the AI recognizes patterns in how a child processes information.
The platform covers fifteen subjects through this unified approach. Financial literacy builds on mathematical foundations the system remembers. Entrepreneurship lessons reference previous discussions about problem-solving approaches. Spanish vocabulary instruction adapts to memorization techniques that worked in other contexts.
State compliance requires quarterly progress reports. These generate automatically from the accumulated memory data, providing detailed documentation of learning progression across all subjects. The reports reflect genuine understanding of each child's development rather than generic assessments.
Building this for my own children keeps the focus practical. Every feature exists because we need it. The persistent memory system emerged from frustration with repetitive explanations and lost context. The multi-subject approach reflects real homeschool requirements, not theoretical completeness.
The technical stack supports this vision: Next.js for responsive interfaces, Supabase for data management, Railway for reliable deployment. Anthropic's Claude provides the reasoning capability, while H.U.N.I.E. supplies the memory persistence that makes long-term learning relationships possible.
Other properties in the Jonomor ecosystem will connect to H.U.N.I.E. over time. Evenfield serves as the proof of concept, demonstrating how persistent memory transforms AI interactions from isolated conversations into ongoing relationships.
The difference between tutoring with and without memory is the difference between meeting a new teacher every day and working with someone who knows your learning history. One requires constant reintroduction. The other builds on established understanding.
That continuity changes everything about how AI can support education. Not through flashier interfaces or more content, but through genuine understanding that persists and deepens over time.
Top comments (0)