oction labs
Company overview
Secure AI deployment and vetted data for regulated industries
Brandon Gill  ·  MJ Dewji  ·  Leonel Garcia  ·  June 2026
Founding stage Mining · Energy · Corrections · Government · Engineering · Education octionagency.com
brandon@octionagency.com  ·  mj@octionagency.com octionagency.com
Foundation
"We deploy AI inside regulated environments, de-identify and vet the data those environments generate, and license the resulting datasets to companies training the next generation of AI models."
1
SECURE AI DEPLOYMENT
Revenue: Day 1
2
DE-IDENTIFICATION PIPELINE
Revenue: Year 1+
3
VETTED DATA LICENSING
Revenue: Year 3+
The origin
We built the infrastructure before we had a client. That was the point.
Every AI company in the world is pitching regulated industries. None of them can actually serve regulated industries.
The reason is always the same: data cannot leave. Cloud AI requires data egress. Regulated environments prohibit it. That is not a negotiation problem. It is a structural problem that no terms-of-service update can fix.
Oction was built around the one model that works: the AI goes inside the perimeter. The data never leaves. The de-identification engine extracts what can be safely used. Compliance is not a feature we added. It is the architecture we started with. That decision meant building before selling. So we did.
WHAT EXISTS TODAY
7-AGENT HIVE
Running 24/7 on sovereign hardware
MISSION CONTROL
Live operational dashboard — mc.oction.agency
SOVEREIGN COMPUTE CLUSTER
Three dedicated hardware nodes, private encrypted mesh
DE-IDENTIFICATION PIPELINE
k-anonymity, full audit log, PIPEDA-ready
1.5ms RETRIEVAL  /  100% EVAL ACCURACY
CAG memory system — production benchmarked
85%+ HALLUCINATION REDUCTION
Internal evaluation — no external dependency
Pre-revenue. Post-build. The only company at this stage with this infrastructure already running.
The problem
Regulated industries cannot legally use general-purpose AI
Mining and natural resources
Geological surveys, Indigenous consultation records, and export-controlled resource data cannot leave secure environments.
Energy and utilities
Pipeline operations data and critical infrastructure specs are blocked from cloud AI by NERC CIP and provincial energy regulations.
Corrections and justice
Inmate records, sentencing data, and parole decisions are locked under strict privacy law. No cloud AI vendor qualifies.
Government and municipal
Provincial data-sovereignty laws and federal procurement rules mandate sovereign infrastructure for any AI-assisted government process.
Engineering and infrastructure
Project blueprints, safety certifications, and liability records fall under export controls and NDA requirements blocking cloud processing.
Education and research
Student records under PIPA/FERPA, proprietary institutional research, and academic IP require on-premise AI with full data residency.
The gapEvery AI vendor pitches "secure." None can prove sovereign infrastructure. Regulated buyers want AI. They cannot legally buy what exists.
Our approach
We build across all regulated sectors, then go deep in 2-3 verticals based on where our first deployments land. Depth makes the data defensible.
What we do
Three layers. Each one compounds the last.
Layer 1
Secure AI deployment
We deploy AI systems inside regulated environments: mining operations, energy infrastructure, corrections facilities, and government. The software runs on the client's own infrastructure or our nodes via secure tunnel. Data does not leave the environment. NIST 800-53, PIPEDA, and provincial data-sovereignty compliance built in.
Revenue: deployment and managed services fees
Layer 2
De-identification and vetting
As the system operates, data interactions are anonymized in real time. We classify every data type by what can and cannot be shared under applicable law. The result is a structured, verified, shareable dataset built from real regulated-industry activity.
Moat: no other party has legal access to this data
Layer 3
Vetted training data licensing
The curated datasets are licensed to AI companies building domain-specific models. Energy sector AI developers, mining intelligence platforms, and justice-system AI vendors all need training data from regulated environments. There is no compliant alternative source.
Revenue: recurring data subscriptions and one-time licenses
The niche strategy: we go deep in a small number of regulated sectors before expanding. Depth makes the data defensible.
Current state
The infrastructure is built. The clients are next.
AI Agent Mesh
Julian Pierce  LEAD AGENT
Co-CEO, Systems and Infrastructure  /  Hermes stack  /  24/7
7 specialized agents running 24/7: systems, operations intelligence, security, research, data, content, and sales. Each agent has a defined role, identity-grounded memory, and access to the shared knowledge vault.
Lucius Atlas Avery Kai Oscar Finn
Mission Control
Operational dashboard running live across private infrastructure. All agent activity, system health, and task coordination managed in real time. Access-controlled via Cloudflare, hosted on our own hardware. Zero third-party dependencies.
mc.oction.agency  /  Cloudflare Access restricted
Sovereign Hardware Nodes
Three dedicated compute nodes running production workloads across a private encrypted mesh. Models hosted locally. No external API dependencies for core operations. This is the exact architecture that gets deployed inside regulated client environments.
Node A  16GB Node B  16GB Node C  8GB
Data Intelligence Pipeline
Edge-first de-identification with k-anonymity and audit logging. Built, tested, and running against internal data flows. Architecture validated before the first client engagement. Six monetization layers designed, Layer 1 operational.
k-anonymity  /  audit trail  /  PIPEDA-ready
1.5ms
CAG RETRIEVAL p50
0.7ms
CACHED RESOLVE p50
94K+
INDEXED CHUNKS
100%
EVAL ACCURACY
85%+
HALLUCINATION CUT
7agents
LIVE 24/7
We are not pitching a concept. We built the operating system first, then we go find the clients it was designed for.
Why now
Three converging forces create a 24-36 month window
01   Regulatory urgency
NIST 800-53 Rev.5, sector-specific AI procurement rules, and EU AI Act spillover are creating mandatory procurement criteria that only sovereign infrastructure can satisfy.
02   Training data scarcity
The next wave of AI model development requires domain-specific, high-quality training data from regulated environments. This data is locked behind compliance walls that no cloud AI company can cross. Oction is purpose-built to get inside first.
03   Talent arbitrage
Compliance-fluent AI engineers are scarce and unrecognized as a hiring category. The incumbents compete on price for generalist talent. We recruit for regulatory depth first. Competing firms have not built this team profile yet.
Miss this window and incumbents (Deloitte, CGI, IBM) will fill it with overpriced, underperforming solutions.
Market opportunity
Total addressable in 5 years: $1.1B
SegmentTAMSAM (5-year)Notes
Regulated AI deployment services$8.4B$420MMining, energy, corrections, government - managed services
AI training data licensing$4.1B$380MVetted, de-identified regulated-sector datasets
Municipal AI infrastructure$2.1B$180MPredictable procurement, defined compliance path
Domain model fine-tuning data$1.8B$120M (Year 3)Niche-specific model weights and curated corpora
$16.4B
COMBINED TAM
$1.1B
ADDRESSABLE (5yr)
$420M
LARGEST SEGMENT
Year 1
REVENUE TARGET
The deployment business funds operations. The data licensing business is where the margin compounds. No competitor is positioned across both.
Our identity
Infrastructure that earns trust before it earns revenue.
Oction is not a product company. We are an infrastructure company. The brand reflects that: precise, dark, and functional. One point of colour, used only where it matters.
Dark by default
We design for engineers, not pitch rooms. Precision over decoration. No gradients. No hype. No exclamation marks.
Sovereign by design
The hardware is ours. The data stays yours. The compliance is built in, not claimed. This is not a feature. It is the architecture.
One signal
Coral is the only accent in our design system. It is used once, where it matters most. That discipline is the same discipline we bring to every deployment.
The pipeline
Secure deployment is the entry point. The data asset is the business.
Year 1: deployment fees. Year 3: data licensing at margin. Year 5: the most defensible regulated-sector dataset in the market.
Depth in a few niches beats breadth across many.
1
Secure deployment
Oction deploys AI inside the regulated environment. All processing stays within the client's infrastructure or our secured nodes. No data egress.
2
Real-time de-identification
Every interaction is anonymized as it occurs. PII and identifying signals are stripped at the point of generation, not after the fact.
3
Shareability classification
Each data type is assessed against applicable law and client consent agreements. A binary determination is made: shareable or not shareable. No grey zone.
4
Vetted dataset licensing
Curated, verified, sector-specific datasets are licensed to AI companies. The buyer gets legally defensible training data from environments they cannot access themselves.
Competitive landscape
No direct competitor for the full stack
CompetitorTheir weaknessOur position
General-purpose cloud AI vendorsCannot sign data-residency agreements. Sensitive institutional data cannot leave the facility. Cloud architecture is the disqualifier.We operate inside the environment. They cannot.
Deloitte / CGI / Accenture$300+/hr labour, no AI expertise, no data pipelineSoftware-first, faster, and we retain the data asset
AI training data brokersPublic web scrapes, synthetic data, no regulated-sector accessReal, verified, de-identified data from regulated deployments. No viable alternative source exists.
Local MSPs and IT vendorsNo AI expertise, no compliance fluency, no de-identification pipelinePurpose-built for compliance from day one
Moat: compliance fluency × inside access × de-identification pipeline = the only source of legally defensible regulated-sector training data at this scale.
Team
Who builds and operates the system
Brandon Gill
Co-Founder, CEO
Leads client acquisition, capital strategy, and government alignment. Responsible for building relationships inside regulated sectors before anyone else was paying attention to them. The commercial bet is his.
MJ Dewji
Co-Founder
Strategic capital deployment and long-range positioning. Oversees how the company allocates resources across the three-layer model as each layer matures. Keeps the compounding logic intact as the company scales.
Leonel Garcia
Partner, COO
Runs the operational engine. Oversees every client deployment from kick-off through go-live and ongoing support. Responsible for delivery consistency as the client roster grows and the infrastructure scales.
Shamir Cheema
Partner, CRO
Builds and runs the revenue pipeline. Responsible for sector penetration strategy, direct outreach to regulated-sector decision makers, and the commercial systems that turn conversations into contracts.
AI staff (active, 24/7)
Lucius - Operations intelligence Atlas - Security and forensics Avery - Research Kai - Data Oscar - Content Finn - Sales
We operate around the clock. We do not burn out. We do not leave for Google.
Our vision
The infrastructure company that owns the most defensible AI training dataset in the market.
We are purpose-built to operate inside the most restricted data environments in the world.
We will de-identify it, vet it, and classify exactly what can leave.
Companies building the next generation of AI models need this data. There is no other legal way to get it.
The deployment business gets us inside. The de-identification pipeline builds the asset. The licensing business is the outcome.

Every niche we go deep on makes the dataset harder to replicate and more valuable to the buyers who need it.
How we get there
Four priorities, executed in sequence
01   Staffing
Build the core team first
Deployment engineers, compliance specialists, and client success operators. Each regulated sector requires dedicated staff who understand the environment deeply. The constraint is people, not opportunity. We hire for niche depth, not breadth.
02   Hardware
Sovereign compute infrastructure
Dedicated nodes that run inside regulated environments. Hardware is not optional in this model. The AI cannot leave the facility. Our infrastructure goes in. This is what physically separates us from every cloud-based competitor.
03   Legal
Compliance and licensing framework
Data licensing agreements, PIPEDA certifications, government procurement qualification, and IP protection across all three layers. The legal structure is what makes the data asset sellable and defensible over time.
04   Advertising
Targeted market entry
High-precision outreach to regulated-sector decision makers in mining, energy, and corrections. This is not mass advertising. It is a small, well-defined buyer pool reached through direct relationship development and sector-specific positioning.
Mining and energy first. Depth in two niches builds a dataset no competitor can replicate. Once established, the model extends to corrections and other regulated sectors on the same infrastructure.
oction labs
Sovereign AI infrastructure for regulated industries. The most defensible data asset in the market, built from day one.
Brandon Gill  ·  brandon@octionagency.com
MJ Dewji  ·  mj@octionagency.com
Leonel Garcia  ·  leonel@octionagency.com
Oction Labs Founding stage · 2026 Founding stage · 2026 octionagency.com