Liminal Technologies Group

Engineering the threshold between intelligence and action

I build ambient AI systems, presence infrastructure, and developer tools that operate at the boundary where technology becomes invisible: running on your hardware, under your control.

See the work → Get in touch
Selected Results
100%
On-Device Inference
4.3M
Barcodes Catalogued
100,646
Wines Referenced
1,295
Screen Lines Recovered

Tests are contracts. Thousands of them run across the portfolio, on every build.

On the bench now: an always-on local assistant, autonomous agents running on my own hardware, and local document comprehension. Details when the filings clear.

Projects & IP Portfolio

Systems built at the edge

Every project under the Liminal umbrella shares one conviction: the most powerful technology is the kind that disappears into your environment. Local inference. Privacy by architecture. Zero cloud dependency.

Active Development

OmniTrade

DAYBREAK

Algorithmic trading engine built on a J-pattern detection system with an 8-state finite state machine. Runs live and fully audited: every fired signal is logged and resolved through the cascade, and the whole session reconstructs from the decision ledger.

J-Pattern Strategy8-State FSMDecision LedgerPython
Active Development

ReceiptTracker

Automated Expense Intelligence

Receipt capture and categorization. Pulls line items, vendors, and totals from photos and PDFs, backed by a 4.3-million-product barcode catalog, and feeds structured data into personal and business accounting.

OCR Pipeline4.3M CatalogMobile ScannerMulti-Entity
Shipped

AMPHORA

Local Cellar Intelligence

A fully local wine cellar tracker. Every bottle runs a lifecycle from ordered to poured, grounded by a 100,646-wine reference database with on-device label reading. Your cellar, your data, your network.

Local Vision100K Reference DBLifecycle
Shipped

FRAMERATE

Video Comprehension

Local video-comprehension pipeline. Transcribes audio, deduplicates frames, and recovers on-screen text the narration never says: 85% of frames pruned, 1,295 lines recovered from a single source. CPU-only, no cloud.

Local TranscriptionOCR RecoveryCPU-Only
Shipped

SHOPLIST

Aisle-Aware Lists

A shopping list that thinks. It auto-categorizes items and orders them by a per-store walk path, so the list lines up with the aisles. Self-hosted, mobile and desktop.

Auto-CategorizationWalk-OrderSelf-Hosted
In Development

PresenceBridge

Pure Mirror Architecture

Real-time status synchronization across platforms with a pure-mirror architecture: any status change propagates everywhere, with no precedence logic and no hierarchy. Multi-calendar sync across M365 and Google. Currently being rebuilt.

Multi-Tenant SyncMS GraphICS / CalDAV
Functional Testing

WakeForge

Custom Wake Word Engine

CLI tool for training and deploying custom wake words. CPU-only inference, Docker-first, ONNX output. Built for edge devices that need always-on voice without a cloud round-trip.

OpenWakeWordONNXDockerCPU-Only
Shipped

LEANPHONE

Device Auditor

Android auditor for battery drain, permission creep, and deadweight apps. It surfaces what is quietly costing you, and why.

Battery AuditPermission AuditAndroid
Functional Testing

CRUCIBLE

CONDUCTOR + PARLEY

Structured multi-agent deliberation. Independent AI instances work through a shared workspace, one driving execution and one driving critical analysis, across dedicated build and deliberate modes.

Multi-AgentShared WorkspaceStructured Deliberation
Architecture Phase

AVATAR

Voice & Visual Synthesis Pipeline

End-to-end synthesized presenter video. Voice synthesis paired with visual generation produces natural conversational avatars from a single reference image and voice sample. Built for personal content production; all training data self-sourced and consented.

XTTS-v2MuseTalkVoice CloneGPU Accelerated

Built at the edge,
not in the cloud

I believe the next generation of intelligent systems won't live in distant data centers. They'll run on your hardware, in your space, under your control. Every tool I build prioritizes local inference, privacy by architecture, and the principle that the best technology is the kind you stop noticing.

My engineering philosophy is simple: all behavioral intelligence lives in deterministic code. Models are commodities, replaceable and benchmarked and never the moat, and the system around them is the product. That makes what I build vendor-proof, auditable, and able to outlast any single model generation.

Local First

If it can run on your hardware, it should. Cloud is a fallback, not a foundation.

Model Agnostic

Swap the model, keep the system. The architecture survives any vendor decision.

Deterministic Core

Intelligence in code, not prompts. Testable, auditable, reproducible behavior.

Privacy by Design

Your data never leaves your network. No telemetry, no training, no exceptions.

The person behind the threshold

Liminal Technologies Group is my IP holding company and development lab. I build at the intersection of AI infrastructure, enterprise operations, and edge computing.

The same discipline that standardizes software across complex organizations is the discipline that builds AI systems engineered to outlast any single model generation. I bring an operator's mindset to every project: ship it, instrument it, make it survive contact with the real world.

My background spans enterprise ERP implementations, automation at scale, decades of hands-on industrial and fabrication work, and AI systems integration. The kind of engineering that only comes from solving real problems in production, not in a lab.

At a glance

Focus
AI Systems & Edge Infrastructure
Background
Enterprise Technology & Solving Problems Others Can't or Won't
Specialties
Local AI, Automation, Systems Integration
Service
U.S. Army Veteran
Based
South Florida
Entity
Liminal Technologies Group LLC

Let's build something

Local AI infrastructure, enterprise rollouts, and automation built to survive vendor churn. Inquiries, partnerships, and interesting problems welcome.

Book a call →

Or reach me directly at [email protected]