Physics informed AI · Digital twins · Real time optimisation
The comfort zone is
the most expensive place
in a process plant.
Avanti builds the operating twin: your columns, reactors and more, optimised across the whole envelope and live in days.
- 5–17%
- energy gap
- 2–11%
- yield gap
- 45 days
- start to live
Redrawn after Lahiri, Profit Maximization Techniques for Operating Chemical Plants (Wiley 2020).
The 7 a.m. report
7 a.m. The overnight report is on the operations manager’s desk.
He knows profit is evaporating — shift by shift, somewhere across his reactors and columns.
Nobody in his plant can tell him exactly where, why, or what to do about it.
What the gap is worth
5–17% energy
2–11% yield
Measured across deployments and literature.
Avanti = your AI process engineer
Your plant gets its own AI process engineer.
A 24/7 process engineer that never leaves the plant — continuously watching, understanding, predicting and optimising your process.
No vacations. No shift changes. No handoffs. No waiting for the morning report.
It recommends. Your team decides.
- Live plant data
- Avanti AI
- Digital twin
- Optimisation
- Recommendation
- Better outcome
The operating loop
- 01
Watch
Continuously reads live process data for deviations and drift.
- 02
Understand
Physics models and plant knowledge explain the numbers.
- 03
Diagnose
Says what changed, why it happened and what it means.
- 04
Simulate
Tests the change on the twin before it reaches the unit.
- 05
Optimize
Balances throughput, quality, energy, yield and constraints.
- 06
Recommend
Delivers a set point your operators can act on.
An AI process engineer that is always on the job.
Not another dashboard, analytics platform or tool your engineers have to remember to use. Your engineers make the decisions — Avanti makes sure they have the intelligence to make better ones.
Recommended action
example dataAdjust reactor temperature
215 °C → 218 °C
Projected impact
+2.8% throughput
−1.4% energy consumption
Reason
Process model indicates improved conversion efficiency within current operating constraints.
Explainable. Auditable. Operator-controlled.
The product
An operating twin for your operations team.
Start with one unit operation, prove the value, and expand from there.
Solutions elsewhere are packaged as
- a tool for your experts,
- an enterprise programme.
They assume capacity most plants don’t have.
We deliver the running twin.
DistillX
separation processesReal time optimisation for separation processes, batch and continuous.
ReactX
reactorsReal time optimisation for reactors, batch and continuous.
Btnol_V001
74.70 % → 81.80 %
FTS197
12.8 → 13.1
FT008
9,604 → 10,100
Integrate, don’t migrate
Proof
The twin told them something they didn’t know.
Real plant data. Measured outcomes. Explainable recommendations.
A plant fought bottom slippage for years by adjusting the column. The twin showed the column was never the problem.
1% → 37%
a 1% feed shift moved slippage by up to 37%†
The driver was feed composition — quantified, with the cascade into the downstream columns.
EO reactor
21% lower cost
production cost, from a 1% selectivity gain†
Single column
$350k / yr
energy saving, measured in production†
Reactor microkinetics
< 2% error
prediction error against measured data†
Many more…
Slippage reduction, reactor stability, golden batch…
† From our live deployments and peer reviewed research. Detailed in our white paper, “Physics enabled AI in Chemical Manufacturing”, available on request.
Our ongoing collaborations
The technology
Physics informed AI models + agentic orchestration
Physics makes the wider search safe. The benchmark says the model is right.
Physics informed AI
01Physics modelling, data generation & validation
First principles models of the process facility. Result quality rises before a single neural net is trained.
02ML modelling + first principles and constraints
A one time hybrid build: machine learning bounded by conservation laws and your unit's real constraints.
03Physics and AI powered recommendations
Set points the plant can actually hold, in your operators' own variables.
v²/2 + ∫ dp̄/ρ(p̄) + Ψ — conservation bounds every prediction
Agentic orchestration
The benchmark
Validated against your own measured data and a commercial process simulator, so the wider search is backed by evidence rather than confidence.
Honest boundary: we do not model engineering changes — we optimise the plant you run.
How it works
From your live plant to a set point your operators can hold.
One continuous loop, running while the process runs.
Physics model + live plant data
A first principles model of the process, continuously reconciled against your historian.
The operating twin
Constrained by physics, the twin extrapolates confidently to set points never visited.
The optimiser
A search over the operating handles your team already sets every shift.
Recommendations
Better conditions, quantified as economic opportunity, in your operators' variables.
Why Avanti
Built by process engineers, for the people who run the plants.
This started in a plant, not a lab. A team that had spent careers operating facilities kept watching the same thing: the units could do more, everyone knew it, and nobody could prove where the line was.
Every part of the twin was designed by someone who has had to defend a set point on a morning call.
DeliberateProcess industry only — until we own it.
PhD
Physics, AI and chemical engineering
Doctorates from leading research universities building the first principles core.
25+ yrs
Facilities operating experience
Careers spent running petrochemical units, not modelling them from a distance.
Peer
Published research
Hybrid physics and machine learning benchmarked against rigorous simulation.
Why teams choose Avanti
- Per unit operation · start with one unit, scale to facilities
- Days, not months · minimal touchpoints to go live
- Edge, on premise or cloud · read only integration with your historian
- OpEx aligned subscription · quarterly billing from the operating budget
Schedule a demo
The opportunity is already inside your plant.
Bring us the unit that is costing you the most. We will build the twin, validate it against your own data and show you what the optimiser finds.