Most engineers ship features. I build the whole machine, and the machine that builds the machine.
full stacks payment rails compliance engines anonymity architectures agent fleets
The short version: a decade of banking-grade fintech (CIBC, Interac-member payment rails), converted into an agentic AI engineering system that lets one person deliver what teams deliver. Everything on this page is live, running software, and every number on it is computed from git history, not claimed.
Now: Application Systems Analyst at an Interac-member fintech · Founder, Toronto 🇨🇦
Say hello: architecture, agentic engineering, fintech → [email protected]
| Platform | What it is | Under the hood |
|---|---|---|
| Licentric | Software licensing and monetization, from license keys to AI-agent tokens. 3 minutes to first validation vs. hours on incumbents | Ed25519 offline-first signed licenses · declarative Stripe→license mapping (zero webhook code) · Python SDK on PyPI · TypeScript |
| LawyerServed | North America's transparent legal directory: 1.47M+ verified lawyer profiles in a double-blind two-sided marketplace | Next.js 16 · tRPC · Drizzle · Supabase RLS · Meilisearch at 1M+ scale · Stripe · per-jurisdiction compliance rule-packs |
| SolveRight | Decision intelligence. 155 decision frameworks run simultaneously with deterministic 0–100 scoring and contradiction detection | Three-phase AI+deterministic hybrid scoring engine · <100ms sensitivity analysis · 7 export formats |
| solvemax | A verdict with receipts: decisions interrogated, scored, red-teamed through 5,000 seeded Monte-Carlo futures, then given a blind second-pass audit | Next.js · multi-model LLM failover chain · Stripe · seeded deterministic simulation |
| ArgusTest | iOS device monitoring and AI-assisted debugging for Claude Code / Cursor users. 100% app-agnostic | MCP integration · HMAC-verified webhooks · SHA-256 error-signature dedup · self-healing daemons |
| ShellJolt | macOS cockpit for AI coding CLIs. Cost, context, usage-limits, and model in one live line; stops OOM freezes before they happen | Rust · Ed25519 licensing · adaptive per-session memory engine · cross-CLI (Claude/Codex/Gemini/Grok) |
| Keralora | Invite-only B2B spice-trade platform, Kerala to Canada, with complete buyer↔seller anonymity | Three-wall anonymity architecture (DAL + Postgres RLS + document redaction) · anonymity test suite in CI |
The Provenance-First Wiki · August 2026 · DOI: 10.5281/zenodo.21862190
Why the LLM wiki pattern breaks past a thousand files, and the five invariants that fix it.
A design paper answering Andrej Karpathy's llm-wiki pattern with the case it cannot survive at scale. 310 catalogued sources across information retrieval, agent memory, archival science, and cognitive psychology. Eight candidate architectures scored under locked weights (pairwise consistency ratio 0.009), a 20,000-run Monte Carlo sensitivity sweep, failure-mode analysis, minimax regret, and an analysis of competing hypotheses. The full decision record is published with the paper, including the section stating when you should ignore it.
A quote is a pointer, not a copy. Almost everything else in the paper is a consequence of that sentence.
The machine that builds the products:
- Multi-agent orchestration doctrine: model-tiered routing (reason/build/research), adaptive concurrency with circuit breakers, orchestrator-gated acceptance where no agent's work merges until independently re-verified
- Scrapos: standalone data-supply engine. 73 scrapers, queue-based pipeline, self-healing, drift-gated auto-redeploy
- SwitchboardOS: control plane for LLM cost, observability, and agent governance
- ops-dashboard: zero-dependency live SDLC monitor that watches every project's agents, tests, and git state simultaneously over SSE. Open source (MIT)
- local-gitlab: fully self-hosted GitLab CE + runner. $0/month CI/CD any project plugs into
- Relay: portable partner-marketing engine (attribution, commissions, payouts) that integrates with any product from zero code upward
- ios-claude-toolkit: 22 battle-tested Claude Code skills for iOS, distilled from 75+ production sessions. Open source (MIT)
- Flowen: iOS money-transfer app. Swift 6, MVVM, 3,349 test methods, 9-stage CI pipeline
- Devrule.ai: enterprise AI-rules middleware that validates code changes against configurable rules before AI tools (Claude Code, Cursor, Copilot) can execute them. Multi-tenant RLS, Stripe billing, 40+ test suites
- Dwellium: trust-first hotel booking platform. NestJS monorepo, 469K lines across web, mobile, and backend, Amadeus + Stripe integration
- estate-network: the machine layer under all of it. One repo, three deployment targets (Mac mini, laptops, router), 10 scheduled jobs, install scripts per target, so no critical automation depends on any one machine being awake
Computed from git history across 21 repositories. Script and methodology published here
(docs/METHODOLOGY.md): every number is checkable.
COCOMO-81 prices this codebase at ~522 person-years. One person shipped it in ~3,172 hours. That multiplier is the point.
The public ledger only. 2014–2023 adds an estimated 5,000+ commits across 20+ enterprise projects (teams of 1 to 37), confidential by contract, much of it in corporate systems that never touched GitHub.
Lines attribute to commit dates: work often lands days after it happens, so active-day counts are a floor and tall bars are batch landings. Aug '26 is a partial month. Every chart on this page is generated from the audit data itself, no third-party widgets.
- ops-dashboard: Ops dashboard: monitors a project's git/tests/agents/lanes, serves at… (2026-09-09)
- AbleVarghese.github.io: Portfolio: ablevarghese.github.io (2026-08-20)
- Provenance-First-Wiki: Why the LLM Wiki pattern breaks past 1,000 files, and the five… (2026-08-10)
- keralora.com: Invite-only B2B spice trade, Kerala → Canada, with three-wall… (2026-08-09)
- shelljolt.com: The cockpit for AI coding CLIs: cost, context, limits, model in one… (2026-08-09)
- argustest.dev: iOS device monitoring + AI-assisted debugging via MCP for Claude Code… (2026-08-09)
Auto-generated 2026-09-13 by update_readme.py. Derived, never hand-edited.
Interac-member fintech, Application Systems Analyst: BASE24 / HP NonStop transaction processing, AI-agent orchestration for internal tooling CIBC (4 yrs), Senior Application Engineer: architected and shipped five banking applications to production; Java/Python; CI/CD with IBM UrbanCode; Scrum Master; led the Drive Smart iOS/Android app team as PM; built an internal application platform and CMS used across Solutions, Development, QA, Build and Production teams StrongBase Capital, securities and options trader: quantitative strategies, Greeks-based risk management Spine Hedge (2021–), AI-powered fintech project: the bridge between my trading years and the platform portfolio above Earlier: Mechanical Engineering @ Carleton · Software Engineering @ Centennial · NASA CanSat & Lunabotics team lead · conference speaker on engineering entrepreneurship
All enterprise, corporate, and client work delivered since 2014 remains private and confidential, by contract and by principle. Everything public here is my own, built outside it. The discipline underneath came from those years.
TDD-first · spec-before-code · structural fixes over patches · one source of truth · evidence before conclusions. Codified as machine-enforced rules my agent fleets follow. Quality is a system property, not a habit.
📫 [email protected] · LinkedIn · X @AbleVeez · Toronto, Canada