Production AI, end to end.

Problem

Education usually assumes learning happened because a student attended, watched, completed, or earned a credential. In serious systems like markets, operations, and finance, performance is measured continuously. Learning systems rarely are. Virtu Prep exists to close that gap.

What it does

Virtu Prep is an AI-powered learning platform built around active practice, real-time feedback, adaptive sequencing, and evidence of actual learner change. It serves schools, microschools, workforce programs, test prep environments, and specialized academies as multi-tenant infrastructure for structured, adaptive, measurable learning.

Architecture

Virtu Prep is a multi-tenant SaaS. The platform integrates several specialized services around the main application, including text-to-speech, phonics processing, and a real-time interaction layer.

The AI assistant is built around specialized fine-tuned agents. Each user flow has its own dedicated agent, coordinated by a central agent that routes intent and assembles responses.

The data architecture is behaviorally grounded. Learner interactions emit dense events that flow through a feature pipeline into a learner state layer maintaining mastery, retention, and misconception risk per learner-concept pair. Policy decisions select instructional content based on predicted learning utility, with experimentation infrastructure for controlled comparisons.

Stack

Backend: Java and Spring. Frontend: React. ML pipeline: Python with a custom corpus, dataset generation, supervised fine-tuning, and a strict evaluation harness. Persistence: relational store for transactional records, append-only event store for behavioral telemetry, model registry for deployment traceability. Container-first deployment in the cloud.

What I architect and ship

I architect Virtu Prep end-to-end and remain hands-on across the codebase. I personally authored the system architecture, the data model, the AI agent design, and the deployment topology. I write production code across the frontend, backend, and ML training pipeline. I'm the most active single contributor in both repositories.

Status

In production across multiple academies and learning environments.

Outcome

Virtu Prep adapts sequencing in real time to individual mastery, retention, and misconception risk. Schools launch full learning pathways quickly. Each learner's path generates an auditable evidence trail of what was practiced, what was retained, and where misconceptions were corrected. The behavioral data captured during interactions feeds back into the platform's instructional intelligence.

What it is

Real Merit Protocol is a public framework I authored for continuous measurement, real-time analysis, and adaptive learning, organized as a structured open protocol at realmeritprotocol.com.

It prescribes how a learning environment should be designed to measure real ability.

Why it exists

Real Merit Protocol is a public proposal for measuring real ability through evidence. Measurable, governed, auditable. Community-built, publicly readable, implementable in pieces.

Author process

I wrote the protocol solo. The core vision is that AI systems must integrate multiple senses, including typing dynamics, response timing, attention markers, audio behavior, and eye and focus signals, to produce better generated content and more accurate measurement of learner ability. The protocol formalizes that vision into implementable sections any operator can adopt.

Integration with Virtu Prep

Real Merit Protocol is operationalized inside Virtu Prep. Every academy generated through School Builder Bot references the Real Merit Protocol by default in its features, comparison, and footer copy, and uses realmeritprotocol.com as the standard privacy reference. The protocol is operationalized as infrastructure in an active production system.

Status

Published and live at realmeritprotocol.com. Operationalized inside Virtu Prep as the standard measurement and privacy infrastructure across all academies generated through School Builder Bot.