Building MicroVault
Notes from the workbench on a microeconomics mastery platform. What I underestimated, what I am fixing, and what I think the next generation of study tools has to look like.
Most online study tools for microeconomics are textbooks with a search bar. They are comprehensive, flat, and built on the assumption that if a student reads the explanation, the student has learned. This is not how microeconomics is actually learned. I built MicroVault because I had used every existing tool and none of them treated the subject the way it deserves to be treated.
What the existing tools get wrong
The first failure is uniformity of weight. Every concept in a textbook is given roughly equal real estate, which is the opposite of how the subject actually works. The rule that a monopolist sets MR equals MC is not the same kind of object as the definition of a budget constraint. One is a mechanism. The other is a frame. Treating them with equal emphasis trains the student to memorize rather than to think.
The second failure is the separation of intuition, formalism, and application. Most platforms give you the math in one place, the graph in another, and the case studies in a third tab if they include them at all. The student is expected to integrate these themselves. They almost never do.
The third failure is passive media. You read. You watch. You take a quiz. You move on. The actual work of understanding microeconomics is in manipulating curves, generating problems, comparing structures. None of which textbook readers do at sufficient volume.
The architecture I chose
MicroVault is built around nine curriculum modules and six interactive instruments. The modules cover the full arc from consumer choice through oligopoly. The instruments are the part that matters more than the curriculum, because the instruments are where the work happens.
The Forge is a problem engine with three modes. Drill mode for fast feedback, Simulation for timed exam-style practice, Adaptive which surfaces the topics the student misses most. The Graph Sandbox treats every curve as manipulable, because that is how the subject is actually tested. The Formula Vault attaches use cases and traps to every formula, not just the equation. Flashcards run spaced repetition across definitions and structural rules. The Concept Map shows how everything connects, which is the view a student usually only gets after the course is over.
Aladdin sits across all of it. Most AI tutors are chatbots that flatter the user. Aladdin was prompted to teach Socratically and to refuse the easy validation that makes chatbots feel productive without actually being so. When a student asks why marginal revenue is below price for a monopolist, Aladdin walks through the mechanism rather than reciting the answer.
What I underestimated
Three things, ordered by how much they hurt.
I underestimated the difficulty of designing problems that are pedagogically useful rather than algorithmically generable. The first version of the Forge could generate unlimited problems. About a quarter of them were trivial, another quarter were ambiguous, and the remaining half were occasionally brilliant and occasionally indistinguishable from the trivial ones. The fix was to template harder, constrain the parameters more aggressively, and accept that fewer good problems beats infinite mediocre ones.
I underestimated how much content quality matters relative to content volume. The first instinct when building a study platform is to fill it with everything. The right instinct is to fill it with less, and to make each piece earn its place. The Primer module is now ten minutes long and does more work than chapters that were three times that length.
I underestimated the cost of running an AI tutor. Aladdin is the most loved feature of the platform and also the line item I watch most closely. The system prompt is now four thousand tokens because that is what teaching well requires. Every conversation is real money. The economics work because the platform is positioned as serious, but the margin is narrower than a casual observer would assume.
What the next version has to fix
Two things that bother me about v1.0 and that v2 will address.
The first is that the Concept Map is currently a beautiful diagram but not yet a study object. Students should be able to mark nodes as mastered, see their progress visually, and have the map adapt. This is the structural feature that would change how the platform feels.
The second is that Aladdin does not yet remember a student across sessions. Right now each conversation starts fresh. A real tutor knows what you struggle with, what you have already covered, and what you keep getting wrong. The infrastructure for this is straightforward. The harder question is what to do with that memory once you have it without making the experience feel surveilled.
The broader point
I built this because the way microeconomics is taught online is not the way it is actually learned. The frameworks for genuine mastery exist. They have just not been productized for a serious student. There is more of this kind of work to do across every academic subject. MicroVault is one node in a bigger thesis I have about study platforms, which I will write about separately.
For now, the platform is live at micro-vault.com. If you are studying microeconomics, the Primer is ten minutes. Start there.