# Renacore — Map of Content A map of the **ideas** behind Renacore — the clearest live example of [[The OT Intelligence Layer]] playing out inside one company. Five ideas stack on top of each other: failure is a forecasting problem, not a budget problem; physics is what keeps the model honest; read-only access is the whole sales motion; a human always signs the work order; and the data you collect is what makes you hard to remove. The company facts come last, on purpose. Start anywhere and follow the wikilinks. > [!abstract] How to read this map > Five layers, bottom to top. A **thesis**: the cost of catching a failure too late dwarfs the cost of upkeep, so the real job is forecasting. A **core idea**: [[Physics Informed Neural Operators|physics-constrained]] prediction, so the model can't be confidently wrong. An **architecture** that doubles as the go-to-market: read-only over the existing plant. A **trust posture**: the agent drafts, a human approves. And a **company**: Qatar LNG-first, pre-revenue, seed-stage. Read top to bottom for the story, or jump to [[#The company, compressed]] for the deal facts. > [!info] At a glance > - **What** — An asset-centric, physics-informed AI that sits *read-only* over existing SCADA/PLC/historian systems and forecasts failure risk on critical LNG rotating equipment 30/60/90 days out. > - **Core idea** — Statistical features + gradient-boosted trees, every prediction validated against physics constraints. Physics as a *guardrail*, not just a loss term. > - **Wedge** — A read-only edge connector at Purdue L3/DMZ, OPC UA, outbound HTTPS 443 only. Integrates in <30 days because nothing in the plant has to change. > - **Market** — LNG / rotating equipment, Qatar & GCC first (QatarEnergy, Shell Qatar, Dolphin, Nakilat), on the back of existing energy relationships. > - **Stage** — Pre-revenue on the LNG product; relationship-stage pipeline. Raising a $3M seed at $12M pre. --- ## 1 · The thesis — failure is a forecasting problem Most plants treat maintenance as a cost line to be trimmed. Renacore thinks that names the wrong problem. > [!quote] The core insight > The pitch is not "spend less on upkeep." It's that one unplanned LNG shutdown costs millions a day. Once you price the downside honestly, both reactive and scheduled maintenance start to look like gambling. The real question is simpler: can you see the failure coming? Put that way, the job is forecasting, and a forecast is something software can own. It's the [[Predictive Maintenance in O&G]] argument aimed at the most expensive assets in the plant — and it's [[The OT Intelligence Layer]] inside one company: software that sits over the control system, reads the telemetry, and turns it into a risk you can act on early. - Cost framing → [[Renacore/The Reactive Maintenance Gap]] · [[Maintenance CapEx]] - Why this is a *layer*, not a point tool → [[Renacore/Asset-Centric Intelligence Layer]] - Where it sits in the value chain → [[Natural Gas Value Chain]] ## 2 · The core idea — physics is the moat The edge isn't more machine learning. It's machine learning boxed in by physics so it can't be confidently wrong. > [!tip] Why physics-constrained, not black-box > On a plant floor, a wrong alert is worse than no alert. It burns operator trust, and once operators stop trusting the alerts they switch the whole thing off. So Renacore runs gradient-boosted trees over the sensor data, but checks every prediction against hard physical limits before it ships. Physics is the guardrail. Think of it as [[Physics Informed Neural Operators]] taken out of the paper and built for a real plant. The outputs are things an engineer can act on, not scores to admire: **failure probability**, **Remaining Useful Life (RUL)**, a health score, **Trip Risk %**, and a **Latent Degradation Index**. Each one answers "what do I do, and by when?" without asking anyone to trust a black box. - Why a generic ML vendor can't cheaply copy this → [[Bespoke Engineering in Industrial AI]] · [[AI era Defensibility]] - Why narrow + deep beats broad + shallow here → [[The Age of Vertical Models]] · [[Where Domain Evals Matter Most]] - How trust is *earned and measured* → [[AI Verification]] ## 3 · Architecture as strategy — read-only is the whole wedge The most important design choice is what Renacore refuses to do: it never touches the control loop. > [!note] The read-only edge connector > A small connector sits at **Purdue L3 / DMZ**, speaks **OPC UA**, and only makes **outbound calls on HTTPS 443**. No inbound ports, no writes, nothing in the process changes. Because you don't have to modify the plant, it installs in **under 30 days** — and the security review gets easy for the same reason the deployment risk does. Here's the part worth sitting with: the limitation is the sales pitch. "We can't break anything because we can't write anything" is the sentence that gets a nervous OT buyer to say yes. That's [[The OT Intelligence Layer]] in practice — intelligence over the stack, never in it — and it's Zero Trust applied to a plant. - The connector & infra detail → [[Renacore/Architecture and Infrastructure]] - The security posture as a sales asset → [[Renacore/Security and Zero Trust]] - Why this is a durable position → [[7 Powers]] · [[Defensibility Principles MOC]] ## 4 · The trust posture — agentic, but human holds the pen Renacore is an agent that never gets the last word. > [!warning] Autonomy stops at the work order > The agent drafts a work order into the CMMS (**SAP PM**, **IBM Maximo**), and a human approves it. It never takes a control action on its own. In a plant where things can catch fire, that's the only version of [[Autonomous Agents]] anyone will buy — and it's why the system gets adopted, not a compromise on it. Nothing goes live on trust alone. A model has to clear a bar first: **>85% precision in backtest**, **fewer than one false alert an hour**, **zero physics violations**, and a trace that explains its reasoning. Clear it and it earns a **Model Release Certificate**. Three checks stand in front of production: input hygiene, a golden-batch core, and a final sanity check on the logic. - The gating system in full → [[Renacore/Physics-First Quality and Governance]] - Why domain-specific evals decide trust here → [[Where Domain Evals Matter Most]] ## 5 · How it sells — the data is the moat, the relationship is the door › The sales idea rhymes with the technical one: the more of the plant you monitor, the harder you are to rip out. > [!info] Land-Prove-Expand, not product-led > No self-serve. Discovery → technical workshop → data review → **a small paid pilot on one or two critical assets** → an ROI report → a multi-year rollout to the whole train, then the fleet. The first bite is small on purpose. The prize is being installed everywhere. Two different things protect Renacore. The relationship gets it in the door — access to Qatar LNG operators that a cold-outreach competitor just can't match. The data keeps it there — every month of history makes the next vendor's job harder. As [[Predictive Maintenance in O&G]] puts it: the moat is the data, and the data comes through the relationship. - The pilot mechanics & KPIs → [[Renacore/Pilot Structure and KPIs]] - The commercial motion → [[Renacore/Business Model and Go-To-Market]] · [[Selling AI MOC]] - Why services-shaped delivery is the wedge → [[AI Eats Services Not Software]] · [[Industrial AI Unit Economics]] - Who actually buys → [[Renacore/Buying Committee and Personas]] ## The company, compressed > [!abstract] The deal facts, in one place > - **Origin** — Founded 2025. Pivoted from a broad enterprise-AI platform into LNG predictive maintenance, riding existing Qatar energy relationships. → [[Renacore/Company Overview and Pivot]] > - **Stage** — Pre-revenue on the LNG product. The pipeline is relationships, not signed pilots. → [[Renacore/Traction and Pipeline]] > - **Raise** — $3M seed at $12M pre. → [[Renacore/Investment Snapshot]] > - **Reality check** — The technology story is strong. There's no traction yet, and that's the whole risk: the bet is that these relationships turn into paid pilots before the cash runs out and before the missing certifications block a procurement gate. See [[3 Hard Truths of Deep Tech Commercialization]]. ## Concept map ```mermaid graph TD THESIS[Cost of too-late: failure is a forecasting problem] --> LAYER[The OT Intelligence Layer] THESIS --> FORECAST[30/60/90-day failure risk] FORECAST --> PHYS[Physics-constrained prediction] PHYS --> GBM[Stat features + GBM] PHYS --> GUARD[Physics guardrails] PHYS --> OUT[RUL / Trip Risk % / Latent Degradation Index] LAYER --> READONLY[Read-only edge connector] READONLY --> PURDUE[Purdue L3 / DMZ · OPC UA · 443 out] READONLY --> WEDGE[<30-day install, zero disruption] OUT --> AGENT[Agentic triage] AGENT --> HITL[Human-in-the-loop work orders] AGENT --> CMMS[SAP PM / IBM Maximo] HITL --> GATE[Model Release Certificate] WEDGE --> GTM[Qatar LNG-first, Land-Prove-Expand] PHYS --> GTM GTM --> MOAT[Data moat + relationship door] class LAYER,PHYS,READONLY,AGENT internal-link; ``` > [!tip] The whole thing in one breath > Renacore treats a machine breaking down as a [[Predictive Maintenance in O&G|forecasting problem]], keeps the forecast honest with [[Physics Informed Neural Operators|physics]], gets into the plant through a [[Renacore/Security and Zero Trust|read-only connector]] that can't break anything, leaves the final call to a [[Autonomous Agents|human]] so a nervous buyer can trust it, and sells it in Qatar first — where the [[The OT Intelligence Layer|data it collects]] is what makes it hard to replace. --- ### Related - [[The OT Intelligence Layer]] - [[Industrial AI MOC]] - [[Predictive Maintenance in O&G]] - [[The Age of Vertical Models]] - [[AI Eats Services Not Software]] - [[Defensibility Principles MOC]] - [[Selling AI MOC]] - [[Natural Gas Value Chain]] - [[History of Qatar]] Tags: #industrial-ai #predictive-maintenance #ot-security #company Links: [[The OT Intelligence Layer]] · [[Industrial AI MOC]] · [[Investing System MoC]]