Flagship · Paragon Energy AI
We are building the AI substrate that renewable energy will run on.
A purpose-built, multi-tier AI platform for distributed energy systems — solar, wind, nuclear, and hydrogen. Edge-native. Self-improving. US-origin. Owned end-to-end.
01 — The Thesis
The renewable industry is operating tomorrow's assets on yesterday's software.
Distributed energy is the most consequential infrastructure build of the next two decades. Solar, wind, nuclear, and hydrogen are not adjacent industries — they are converging into a single grid layer that will be operated by software. That software does not yet exist.
The incumbent monitoring platforms are dashboards. They visualize problems. They do not solve them. Multi-manufacturer fleets generate fragmented telemetry across half a dozen proprietary protocols, and the operator's job is still to read alarms, dispatch trucks, and write tickets. The labor cost of simply understanding what is broken — let alone fixing it — is the binding constraint on renewable economics.
Paragon Energy AI is the operating system this industry has been waiting for. We unify multi-OEM telemetry into a single schema, run AI diagnostics at the edge for real-time response, escalate to cloud reasoning for complex cases, and route to third-party models only as a fallback. The platform self-heals where it can, explains itself in plain language where it can't, and learns from every incident across every fleet. That last point is the entire game: every additional system on the platform makes every other system on the platform smarter.
02 — What We've Built
A purpose-built platform for distributed energy operations.
The Platform
A unified operating layer for distributed energy systems across solar, wind, nuclear, and hydrogen. Multi-OEM telemetry ingestion, real-time anomaly detection, incident orchestration, ticketing, and a knowledge base that learns. Built from the ground up for the operational economics of renewable infrastructure — not adapted from generic monitoring software.
What it covers:
- Unified telemetry schema across heterogeneous equipment
- Statistical anomaly detection at the edge
- Detection model lifecycle with progressive deployment
- Incident engine with automated root-cause analysis
- Knowledge base with semantic search and retrieval-augmented diagnostics
Sentinel
The autonomous edge agent that lives on every device in the field. Sentinel runs locally on Jetson-class hardware, ingests telemetry directly from inverters and balance-of-system components, classifies anomalies in real time, and acts — without waiting for the cloud. Offline-capable by design. Sub-100ms diagnostic latency. Self-healing where the playbook permits; intelligent escalation when it doesn't.
What makes it different:
- On-device inference; no round trip to the cloud for routine diagnostics
- Data sovereignty: production telemetry stays on Paragon infrastructure
- Continuous learning loop: every incident teaches the fleet
The Paragon LLM
A large language model purpose-built for distributed energy. Most AI deployments in this industry are general-purpose models with energy-flavored prompts. Paragon is the inverse: a US-origin, permissively-licensed foundation model fine-tuned exclusively on renewable energy telemetry, fault libraries, multi-OEM service histories, and the proprietary corpus of incident-resolution pairs that the platform generates daily.
The longer the fleet runs, the better the model gets. The bigger the fleet, the faster it learns.
03 — Architecture
Edge. Cloud. Fallback. No single point of failure.
Paragon's AI runs on a three-tier inference architecture, with each tier optimized for a different combination of latency, complexity, and availability. Eighty percent of inference happens on the edge, on hardware the customer already owns. The remaining workload routes intelligently based on confidence scoring.
Tier 1 — Edge
On-device inference on Jetson-class hardware. Distilled domain-specific model. Real-time anomaly classification, known-pattern diagnostics, offline-capable operations. Sub-100ms latency. Data never leaves the device for routine work.
Tier 2 — Cloud
Fine-tuned proprietary model hosted on Paragon infrastructure. Handles natural-language operator queries, root-cause narrative generation, retrieval-augmented diagnostics, and complex multi-variable reasoning. Fixed monthly cost; does not scale linearly with fleet size.
Tier 3 — Fallback
Third-party frontier-model API for edge cases and quality benchmarking. Always available, always optional. The strategic posture is to use the best available external models for hard cases while owning the proprietary model that handles the rest.
This architecture also serves as the training data pipeline for the proprietary model. Every Tier 2 and Tier 3 inference contributes to the dataset that improves Tier 1 and Tier 2. The platform compounds with use.
04 — Defensibility
What we own that nobody else can replicate.
The proprietary training dataset.
The most valuable AI asset is not the model architecture — that is open-source. The asset is the dataset. Paragon accumulates structured training data from every incident, resolution, anomaly explanation, and operator interaction across the fleet. After 24 months of operation, the dataset is the moat. A competitor with a better model and no data still loses; we have a worse model and the data, and the gap closes monthly.
The continuous-learning architecture.
Paragon is built to improve. Production data flows automatically into curated training datasets. The model retrains on a scheduled cadence. New versions are A/B tested in staging and progressively deployed to the fleet through a controlled rollout queue. Most platforms ship a model once and watch it decay; Paragon's model gets better every quarter by design.
The multi-energy generalization.
Paragon is the only operations platform we know of training a single AI substrate across solar, wind, nuclear, and hydrogen. The cross-domain transfer learning is itself patentable. Each new energy type added to the platform makes the existing types diagnosed more accurately, because the underlying physical and operational patterns rhyme across domains in ways general-purpose models do not see.
The US-origin technology stack.
All AI components are sourced from US-based entities — model foundations, inference runtimes, vector databases, embedding models, infrastructure providers. This is a deliberate architectural choice driven by NERC CIP requirements for the grid, NRC requirements for nuclear, utility customer trust expectations, and government contract eligibility. Many competitors will spend the next three years retrofitting for these constraints. Paragon was built inside them from day one.
05 — IP
Built to be patentable from the first commit.
Paragon's IP strategy is structured around four categories of protected asset: the proprietary training dataset (trade secret), the fine-tuned model weights (trade secret and patent), the training methodology (patent), and the three-tier inference pipeline with continuous-learning orchestration (patent). The architecture, the data flows, and the corporate structure were all designed in advance to support patent applications, not retrofitted afterward.
A multi-energy AI diagnostic method. A three-tier inference pipeline with confidence-based routing. A continuous-learning system that curates production data into structured training sets and retrains without service interruption. A detection-model lifecycle in which AI-generated proposals are reviewed, simulated, canaried, and deployed back to the edge.
These are not concepts. They are running code.
06 — Partnerships
Anchor partner: Radiant Energy Capital.
Paragon's anchor commercial partner for renewable energy project finance and deployment. Radiant brings the asset pipeline; Paragon brings the operating layer.
07 — The Company
Where Paragon stands today.
- Founded
- October 2024
- Headquarters
- Midway, Utah
- Founder & CEO
- Thomas J. Hanks
- Structure
- Multi-entity stack designed for ITC/PTC transferability and patentable IP separation
- Anchor partnership
- Radiant Energy Capital
- Stage
- Building under anchor partnership; selective strategic conversations
We are at the beginning of a long compound.