This dossier is the proof, not the pitch. It opens with a live, revenue-earning production platform — designed and shipped by one architect with AI governed under strict TDD, security, and quality gates — and then exposes the architecture depth and applied-AI research underneath it. The flagship demonstrates the deliverable: a secure, highly-available, enterprise-grade system running active-active across two availability zones. The exhibit grid is the credibility spine — reproducible benchmark programs, a foundational data-modeling framework, and production AI infrastructure that show the thirty-plus years of enterprise architecture standing behind every line. Every figure is traceable; every claim is defensible.
WaveMAX is a live, revenue-earning, multi-role production platform — administrators, operators, affiliates, and customers across a laundry pickup-delivery network — designed and shipped solo, with AI as a governed pair-programmer under strict TDD, security, and Lighthouse gates. It runs active-active across two Phoenix availability zones on Oracle Autonomous Database behind a Cloudflare global load balancer, with defense-in-depth security and front-end pages measured at 99 mobile / 98 desktop. It is the working proof of the value proposition: secure, highly-available, enterprise-grade systems delivered without the enterprise price tag.
Every user-facing page is held to a Lighthouse-100 release gate across Performance, Accessibility, Best Practices, and SEO, measured on mobile and desktop before ship. The reference site scores 99 / 98 against a comparable vendor platform's ~59 / 83 (measured 2026-05-25). The engineering shows up in the payload: 0.98 MB page weight versus ~4.5 MB, and a mobile Largest Contentful Paint of 1.8s versus 14.4s — an order-of-magnitude difference a real user feels on first load.
Security is built as eight discrete, testable controls rather than a single perimeter: PBKDF2-SHA512 password hashing (100k iterations, per-user salt, constant-time compare); AES-256-GCM encryption at rest on sensitive fields; short-lived JWT access tokens with rotating refresh tokens and a logout blacklist; distributed rate-limiting; CSRF protection; input sanitization against injection and XSS; strict CSP v2 with per-request nonces and no inline scripts; and audit logging to a separate retention stream. The implementation state has been independently verified against the codebase — surfaced strictly as security engineering.
The codebase is governed, not just written. Roughly 142 test files carry a ~2:1 test-to-code ratio, with an integration-first bias that exercises full request-to-database seams. Red-green-refactor is enforced on every non-trivial change; no production code lands without a regression test. Structural rules hold the line over time: an 800-line file ceiling, zero circular dependencies, logger-only output, and a full suite that runs clean without forced exit.
Each exhibit is a repository, a benchmark, or a working system — the depth that makes enterprise-grade delivery by one person credible. Competitor comparisons are kept generic; every figure traces to a reproducible source.
Reproducible benchmark pitting a converged database's optimizer-driven plans against a specialized engine's pipeline executor across 13 scenarios on matched, compute-capped infrastructure.
Eight-way schema-validation benchmark with a subprocess driver protocol and a correctness gate requiring all engines to agree before any timing is counted.
In-process benchmark isolating the algorithmic complexity of binary JSON field access, server-side updates, and join strategies, instrumented with workload-repository capture.
Java library translating a specialized engine's aggregation pipelines into equivalent SQL/JSON, enabling gradual migration without rewriting application logic.
Runnable proofs repository executing every technical claim from the converged-database article series against a free container, nightly, across five access patterns on one seeded dataset.
Foundational framework establishing that documents, graphs, time series, and relational data are complementary projections of one canonical form — replacing polyglot persistence with a single-database architecture.
Open-source visual designer that operationalizes UMT: connect to a live database, reverse-engineer the ERD, declaratively map relational tables to document shapes, and generate and deploy bidirectional duality views.
A production MCP (Model Context Protocol) server that gives AI coding agents durable memory across session restarts and compaction boundaries — saving, resuming, and replaying state, tool calls, and task history. Self-hostable; the infrastructure layer that makes long-running, AI-driven development reliable.
A full-stack tool that turns raw customer documents into sizing, TCO, and executive-summary deliverables under two hard invariants: a deterministic engine computes every authoritative number (the model writes prose, never figures), and a fail-closed local anonymizer requires human review before any data leaves the machine.
A hands-on enterprise RAG reference implementation and curriculum: vector-search fundamentals plus a complete retrieval-augmented agent — ingestion, overlap chunking, in-database embeddings, vector indexing, similarity retrieval, generative-AI chat, and conversation history with out-of-domain handling.