Waste-to-Chemicals Has a Translation Problem, Not a Technology Problem

Waste-to-chemicals does not have a technology problem anymore. It has a translation problem.

The catalysts exist. The sorbents exist. The control models exist. What is genuinely rare is the person who understands both sides well enough to connect them.

Two rooms, two languages

We spend a lot of time in two rooms that rarely speak to each other.

In the first room sit process engineers who have spent twenty years reading a reactor by feel. Give them a model they cannot inspect and they will not trust it. That is a fair position. Industrial plants have been burned before by software that promised optimisation and delivered alarms.

In the second room sit strong machine learning engineers. They can build a control policy in a week. Many of them have never seen what a poisoned catalyst bed does to a plant schedule. That is equally fair. Nothing in their training required it.

Neither room is wrong. Neither room is complete.

What breaks when they don't talk

Two failure modes, both expensive, both predictable.

A control layer that does not understand catalyst chemistry optimises the wrong variable beautifully. The numbers improve. The reactor does not. Tar breakthrough, sorbent exhaustion and pressure drop are all measurable, and none of them are visible to a model that was never told what a catalyst bed actually costs to replace.

A purification system designed without control in mind hits its spec on a test bench and misses it in production. Steady-state performance is the easy case. Syngas from real feedstock is not steady state — composition, moisture and contaminants all move, and a fixed setpoint does not move with them.

There is a third failure mode, quieter than both: three purification stages, each optimised against its own KPI, nothing optimised against the system. The tar reformer chases tar. The sorbent bed chases sulfur. The filter chases particulate. Every stage succeeds. The gas still misses spec.

Integration as an engineering decision

That is the choice we made in the SyngaPure design phase. Build three independent stages, each optimised for its own target — or build one integrated system, optimised at the platform level.

The first option is simpler to procure and easier to explain. The second option is harder, and it is the one that works. Every stage affects every other stage. When feedstock changes, the AI control layer has to see all three at once and adjust them together. When one stage struggles, the others compensate. That is what makes consistent purity achievable across feedstock variability that would break a sequential system.

Not better catalysts. Not better sorbents. Better integration — and the control logic that makes it possible.

The hybrid position

So we build with people who sit in the overlap, or who are willing to learn the other half.

That is a hiring principle and a partnership principle at the same time. On the process side it means engineers who will let a model into the loop, provided they can inspect it. On the control side it means applied AI people who want to understand catalyst chemistry, sorbent regeneration and filtration pressures before they write a policy. Neither half is a shortcut. Both halves are learnable.

Where we are

SyngaPure is a modular gas cleaning platform that integrates catalytic tar reforming, sorptive pollutant binding and advanced particle filtration — including particulate removal below 10 mg/Nm³ — under a locally installed AI control layer that runs on site, not in the cloud.

Development is backed by the BMWK go-inno programme, with research partnerships active on catalyst and sorbent work. That short list of partners is deliberate: catalysis, materials, AI control, sensing, integration. Each one closes a gap we would rather not close alone.

If you sit in both worlds

Process engineering and applied AI, in one person, or in one team willing to build the bridge. If that describes you, we should be talking.

www.zero-x.co