GOVERNED RESERVOIR SIMULATION POWERED BY AI
An agentic system that carries the full reservoir-simulation workflow — initialization, history matching, prediction, uncertainty evaluation — inside your standards, inside reservoir physics, and inside your chain of accountability. It proposes every change with a mechanism, knows when a parameter has stopped paying, and asks permission before widening the search.
Not autonomous curve-fitting. Your simulator remains the numerical authority. Your senior engineers remain accountable. The system makes the reasoning visible, auditable — and learnable.
Every one of these models honors the same history. They disagree only about the future — and that disagreement is the real answer. A single "good match" hides it; an ensemble states it.
That is why the workflow is built as a Bayesian ensemble filter — a disciplined family of models conditioned on the same data — not a search for one flattering curve. Predictions come with a defensible range, and the range itself is a deliverable.
Single cases as scouts
Deterministic single-case runs are not discarded — they are put to work. The system proposes targeted single cases to frame the ensemble: probe mechanisms, expose sensitivities, and set physically defensible parameter ranges before the ensemble stage begins. Single case and ensemble are one loop, not two products.
Every cycle is deliberate, recorded, and accountable. Nothing is tried "to see what happens."
The agent never edits a parameter silently. Each proposal names the mechanism, the expected effect, and the data it is meant to reconcile — then the prediction is checked against what actually happened after the run.
Each run passes a battery of independent checks: did the change produce its expected effect, did anything regress elsewhere, what exactly changed in the deck, is this move a repeat of one already tried. A pretty misfit number alone convinces nobody here.
A campaign memory holds every move, its rationale and its verdict. An anti-repeat gate blocks equivalent moves before they are ever run. Sensitivities accumulate across the campaign instead of being rediscovered.
When changes to the working parameter set stop improving the match, the agent says so. It does not quietly widen the search — it requests permission to open other parameters, with the evidence for why. Escalation is a feature, not a failure.
Speed is easy to buy. Discipline is what makes reservoir simulation defensible.
Parameter ranges and mechanisms are anchored to what is known about the specific reservoir — not to whatever makes a curve bend. Every move is cross-checked against the independent evidence:
A model that matches history at the price of unphysical properties is rejected — however good the misfit looks.
Every oil & gas company already has internal standards for simulation studies — initialization QC, history-matching acceptance criteria, prediction setup, uncertainty evaluation. ResSimIQ encodes them as gates in the workflow, not documents on a shelf.
A model does not advance to the next stage until the current stage's criteria are met. A posterior carries an explicit status — provisional or release-grade — never just "done."
The result: every number that reaches a decision meeting can be traced back through the standards it passed.
The system does the work — and builds the people who own it.
Traditional training is bottom-up: years of theory and fragments of studies before an engineer ever owns a full one. The ResSimIQ model is top-down. A young engineer starts at the level of the whole workflow from day one — framing the problem, reviewing the agent's proposed mechanisms, accepting or challenging each move — while the system carries the mechanics underneath.
Senior judgment is not replaced. It is scaled: encoded in the gates, the evaluators and the physics boundaries that every campaign runs inside — and exercised at review, where it belongs.
Capability that stays in-house, built on your own fields — and keeps working long after any external consultant has left.
The same work, tracked at the level each role actually needs — with nothing lost between levels.
The full reasoning trail: every proposal, its mechanism, the run, the evaluators' verdicts, the campaign memory. A working environment for the study — nothing summarized away.
Model status across the asset: workflow stage, match quality, posterior maturity — and, critically, where the agent is waiting on a human decision or has requested to widen the parameter set.
The portfolio view: which models are release-grade, which forecasts rest on what basis, and a complete audit trail behind any number. No black box between the slide and the simulator run.
Human-in-the-loop gates span the workflow from problem framing to release-grade posterior acceptance — each with a defined owner. Approval is a role, not a button.
Your organization decides how much the agent does on its own, per stage: from "proposes only, humans execute everything" up to supervised campaigns. Autonomy is configured, earned and revocable — never assumed.
ResSimIQ connects to your simulator; it does not replace it. The simulator remains the numerical authority — every result comes from your engine, on your models. A pluggable parser architecture reads native result formats where available and standard exports where required, so the system fits corporate IT reality instead of fighting it. Your decks stay your decks: every change is a tracked, reviewable deck diff.
Models, decks, results and campaign records stay inside your infrastructure. Reservoir Mentor is designed to run where your data already lives — your subsurface stays yours.
For organizations standardized on Microsoft 365, review, approvals and reporting can flow through Teams, SharePoint and Adaptive Cards — governance in the tools your gate-keepers already use, while the engineering core stays independent of any single ecosystem.
Tell us who you are and what you run. We'll come back with a technical conversation, not a generic sales deck.
Useful to mention: your simulation stack, the kind of models you fight with, and whether your interest is technical, organizational — or both.
team@ressimiq.com