Too much to watch
Thousands of tags, alarms and trends compete for limited engineering attention. Important process drift can develop long before someone has time to investigate it.
Sentinel · AI process engineer
An engineer that
never stops watching.
Never stops learning.
Sentinel works alongside operations and engineering teams 24/7—continuously understanding process behaviour, investigating deviations, identifying optimisation opportunities and helping determine what should happen next. Underneath, Veldar combines live plant data with first-principles engineering models so Sentinel can reason about the physical process, not historical patterns alone.
Feed
System
System
Process
Unit A
Unit A
Pump
P-204
P-204
Utility
Loop
Loop
Separator
S-17
Continuous engineering watch · 1,248 live signals
S-17
State update Pump P-204 cavitation margin decreasing
Investigation Mass balance reconciled across Unit A
Constraint Separator S-17 approaching hydraulic limit
Decision Alternative operating plan evaluated
Autonomy Approved optimisation envelope active
Edge sync Site model healthy · latency 11ms
State update Pump P-204 cavitation margin decreasing
Investigation Mass balance reconciled across Unit A
Constraint Separator S-17 approaching hydraulic limit
Decision Alternative operating plan evaluated
Autonomy Approved optimisation envelope active
Edge sync Site model healthy · latency 11ms
01 · WatchContinuously monitor the process
02 · InvestigateUnderstand what changed and why
03 · RecommendEvaluate actions and trade-offs
04 · LearnTrack outcomes and improve decisions
The engineering capacity problem
Your plant generates more data than any engineer can continuously investigate.
Sentinel does not stop watching.
Sentinel does not stop watching.
Investigations are reactive
Engineers often begin analysis after an alarm, production loss or equipment issue appears—when value may already have been lost.
Knowledge does not scale
The best operational decisions depend on experienced engineers combining process history, equipment knowledge and first-principles reasoning. That expertise is difficult to apply everywhere, all the time.
How the AI engineer works
From plant evidence
to engineering judgement.
Sentinel follows the same loop as a strong process engineer: gather evidence, build context, test explanations, evaluate options and learn from outcomes. Physics and engineering models sit underneath that workflow, giving the AI a grounded understanding of how the process should behave.
Gather evidence
SCADA, historians, PLCs, laboratory data, CMMS, edge devices and engineering records.
Build context
Relate signals to assets, process streams, operating modes, design intent, constraints and dependencies.
Investigate
Reconcile data quality, infer hidden physical state and form competing explanations for process behaviour.
Test hypotheses
Use mass and energy balances, fluid, thermal, mechanical and process-specific models to test what is physically plausible.
Recommend
Evaluate scenarios and rank actions by expected impact, confidence, risk and operating constraints.
Learn from outcomes
Capture what operators did, what happened next and how future engineering decisions should improve.
Engineering reasoning you can inspect
Evidence. Hypothesis.
Recommendation. Outcome.
Sentinel shows the engineering case behind every recommendation.
Every investigation retains its evidence: observed signals, inferred state, expected physical behaviour, competing causes, uncertainty, recommended action and eventual outcome. Engineers can inspect the case before deciding whether to act.
This is the foundation for trust: not a black-box score, but an auditable engineering judgement.
ObservedDischarge pressure
Current value6.42 bar
Estimated stateCavitation margin
StatusNarrowing
Expected stateAt current duty
Deviation−11.8%
Likely causeInlet condition change
Confidence86%
Recommended actionReduce speed 3%
Constraint checkApproved
What your AI process engineer does
Engineering work,
performed continuously.
Sentinel is designed to take on the repetitive investigative work that consumes engineering attention—while keeping people in control of consequential decisions.
Continuous process surveillance
Track the process across assets, streams, constraints and operating modes without waiting for an engineer to open a dashboard.
- Persistent plant watch
- State estimation
- Data quality and uncertainty
Explain what changed and why
Form and test root-cause hypotheses using plant evidence, engineering context and first-principles behaviour.
- Root-cause hypotheses
- Operating-envelope analysis
- Cross-asset effects
Evaluate what to do next
Test available actions before they are applied and rank them by expected impact, confidence, constraints and operational risk.
- What-if analysis
- Constraint checking
- Expected-impact ranking
Work alongside your team
Deliver engineering recommendations into existing workflows with approvals, traceability and clear supporting evidence.
- Role-based decision queues
- Operator acknowledgement
- Engineering handoff
Remember what happened next
Capture decisions, interventions and outcomes so the engineering context becomes richer with every operating cycle.
- Decision history
- Outcome capture
- Continuous model refinement
Progress toward bounded autonomy
As confidence grows, approved actions can move from recommendation to execution within explicit physical, safety and business constraints.
- Approved action envelopes
- Human override
- Full decision trace
From engineering assistant to autonomous operation
Trust is earned.
Autonomy follows.
Sentinel begins by working alongside engineers, not replacing control or safety systems. As its engineering judgement is validated against real outcomes, organisations can progressively authorise more actions inside defined operating boundaries.
Observe
Unify live signals and engineering context into a common operating picture.
Explain
Identify physical causes, deviations and developing process conditions.
Recommend
Propose actions with evidence, confidence and expected operational effect.
Approve & execute
Trigger actions after human approval through existing control and workflow systems.
Autonomously optimise
Continuously act inside validated constraints, with oversight, traceability and override.
Brownfield-first deployment
Give Sentinel one engineering job.
Expand from there.
Sentinel complements the tools your engineers already use—SCADA, historians, DCS, simulations, maintenance systems and existing analytics. It can begin read-only on one repeatable engineering problem and expand without a rip-and-replace programme.
Give Sentinel one job
Select one costly, repeatable engineering problem with available data and a measurable operational outcome.
Build the engineering context
Connect relevant plant data and map assets, process relationships, design intent and approved operating constraints.
Run alongside your engineers
Compare Sentinel’s investigations and recommendations with engineering judgement and actual plant outcomes.
Give it more responsibility
Add engineering tasks, assets, process units and sites while preserving shared context, model knowledge and decision history.
Design partner programme
What would you ask another process engineer to solve?
We are working with a small number of continuous-process organisations to validate Sentinel against real engineering work. The strongest starting point is a repeatable problem where your team already has data, but still spends significant engineering time understanding what is happening, why it is happening and what to do next.
One site or process unit
One repeatable engineering job
Existing plant data first
Run alongside your engineers
Measure the operational outcome
Expand responsibility as trust grows