Physics before models
A learned model may choose inside an envelope that first principles define. It may not invent a setpoint the physics rejects.
About
Paper is the invisible substrate. It should not be run on memory.
Packaging, tissue, print and specialty grades are made on the largest, fastest continuous machines on earth — and the knowledge that keeps them running well lives in a small number of people who are retiring. Pulpum exists to put that knowledge into a system that never forgets a shift.
Live example: Grade change PM4 · 135 gsm kraftliner → 110 gsm testliner, no break, ≤14 min off-spec
Why
Why we are building this
A paper machine is a two-hundred-metre continuous process running a fibre web at up to 2,000 metres per minute. It is unforgiving in a specific way: a stock-prep drift, a wet-end chemistry upset, a forming error or a moisture-profile excursion does not fail loudly — it quietly produces broke, off-spec reels and, eventually, a web break that costs hours.
The industry's answer has been craft. Paper-makers and process engineers who have run the same machine for twenty years know what the sheet is about to do. That knowledge is real, valuable and almost entirely undocumented — and it is walking out of the mill. Meanwhile packaging and tissue demand keeps rising, energy and carbon targets keep tightening, and the plastics-to-paper shift adds grades nobody has run before.
Automation has not closed the gap because base-layer control was never designed to reason. A PID loop holds a setpoint; it does not decide that the steam schedule should be re-phased forty seconds before the basis-weight ramp because the dryer section will unload faster than the header can follow. That is a planning problem, a simulation problem and a perception problem at the same time — which is exactly what an agent system with a GPU at the machine and a twin behind it is for.
So we are building the operations layer: agents that perceive the web, plan the run, rehearse it on an as-run twin, act inside limits the mill sets, and write down everything they did. Not to replace the paper-maker, but to make sure the machine still runs like the best shift on the worst night.
Principles
How we build
Six commitments that shape every decision.
The log is the product
Trust is built by being scrutinisable, not by being confident. Everything is recorded, including the recommendations the crew rejected.
Fail to the existing control
Any loss of inputs, models or policy means we stop writing and hand back. Degrading to today's way of running is always acceptable.
Speak mill
Freeness, kN/m, bar, g/m², CSF. If a recommendation cannot be stated in the units a machine tender uses, it is not ready.
Autonomy is earned per site
Nobody gets unattended operation because a demo went well. Each site sets the level and can lower it at any time.
Say what is aspirational
We mark design targets as design targets. Pre-launch companies that blur that line poison the well for the whole category.
Run timeline
What we are actually building
Not a dashboard. A run engine that executes, explains and records industrial work.
run_8f21c4
PM4 · 6.8 m trim · 1,180 m/min
SUCCEEDED
-
01
ingest.grade_target
SUCCEEDED
0.8 s
Mill Orchestrator — Pulled the 110 gsm testliner spec, customer tolerances and the standing energy budget from mill MES; locked the target envelope for the run.
-
02
twin.simulate
SUCCEEDED
34 s
Mill Orchestrator — Simulated 48 candidate transition recipes on the as-run paper-machine twin — forming, press, dryer and calender — and ranked them on off-spec tonnes, break risk and steam.
-
03
stock.refine
SUCCEEDED
2 m 10 s
Pulp-and-Stock — Stepped refiner specific edge load 1.9 → 1.4 Ws/m and pushed freeness toward 412 CSF while consistency held at 3.4%.
-
04
wetend.dose
SUCCEEDED
1 m 26 s
Wetend-and-Chemistry — Retention aid trimmed to 214 g/t and sizing to 1.1 kg/t against live charge and turbidity; first-pass retention recovered to 78% inside 90 seconds.
-
05
form.headbox
SUCCEEDED
1 m 05 s
Form-and-Press — Re-cut the slice profile across 78 actuators and set jet-to-wire to 0.994 to hold formation index through the basis-weight ramp.
-
06
press.nip
SUCCEEDED
48 s
Form-and-Press — Nip load reduced 620 → 540 kN/m to protect the lighter web; post-press dryness landed at 47.1%.
-
07
dry.steam
SUCCEEDED
3 m 18 s
Dry-and-Coat — Re-phased the steam schedule across 6 dryer groups and rebalanced the hood; reel moisture converged to 7.4% ±0.19 2σ at 6.1% less steam than the standing recipe.
-
08
inspect.web
SUCCEEDED
continuous
Defect-and-Inspect — 18 line-scan cameras streaming; two edge-crack precursors detected at the drive side and cleared by a 40 kN/m nip trim before either propagated.
-
09
approve.human
APPROVAL
52 s
Mill Orchestrator — Speed ramp 1,180 → 1,245 m/min exceeded the site autonomy threshold. Held for the machine tender; approved by J. Okonkwo at 04:57:12.
-
10
reel.qualify
SUCCEEDED
1 m 31 s
Quality-and-Conformance — Reel R-24188 released: full genealogy written (furnish, chemistry, CD profiles, defect map, operator actions) and checked against grade spec.
Focus
What kind of company this is
Pulpum is an independent, pre-IPO startup building original AI and industrial-autonomy software for pulp and paper mills. We are not a consultancy, a systems integrator, a reseller or a division of an incumbent automation vendor, and we do not intend to become one. Revenue is recurring software revenue.
Incorporation as a registered legal entity, a business-domain email address and the live pulpum.com site are prerequisites we are treating as [ASPIRATIONAL] pre-application milestones for the NVIDIA Inception programme, alongside a founding team that includes engineers who have shipped production perception and control systems.
The product category is physical AI: mill-edge perception, reasoning, simulation, optimisation and guarded machine and chemistry action. It is GPU-essential work — high-frame-rate web inspection, CFD and drying simulation, and large-scale optimisation are not workloads that run on a mill server.
Accelerated computing
Physical AI, at the machine
Paper moves at 1,200 metres a minute. Perception has to be local, deterministic and fast, so Pulpum runs GPU inference at the mill edge and keeps training, simulation and optimisation in the cloud or on-prem.
Edge perception
8–24 synchronised line-scan cameras per machine. Sub-100 ms defect classification, sub-250 ms break-risk updates. [ASPIRATIONAL design targets]
Model serving
Vision, time-series prognostics, grade embeddings and process reasoning behind one orchestrator, with deterministic rollback by model version.
Training
Defect vision, break precursors, wet-end response and drying models trained on reel genealogy, QCS histories and operator corrections.
Digital twin
GPU-accelerated CFD, drying and web-dynamics simulation of the as-run machine — 10–100 candidate recipes evaluated per grade change.
Synthetic data
Rare break, contamination, wrinkle and formation-upset variants generated and always validated against real mill events before promotion.
Optimisation
Grade-change sequencing, dryer energy allocation, machine-speed balancing and maintenance windows under production and energy constraints.
Measured
The market
Pulp and paper automation, QCS, web inspection, stock-prep and drying control, mill MES and factory software.
$19B
Total addressable market
$4.7B
Serviceable addressable market
$280M
Three-year obtainable market
~10%
Category growth rate
Figures are design targets and pilot-scoped results [ASPIRATIONAL]. Every number is reproduced from the mill's own reel genealogy, not our telemetry.
Team
Who is building it
A deliberately small team of people who have shipped perception and control into environments where mistakes are expensive. [ASPIRATIONAL — team in formation]
Product & mills
Owns the mill relationships, the pilot programme and the commercial model. Spends more time in control rooms than in the office.
Platform & autonomy
Owns the orchestrator, the policy engine and the audit architecture. Previously shipped safety-critical control software.
Edge vision
Owns the camera pipeline, defect classification and break-precursor models running on the mill-edge GPU.
Papermaking science
A paper-maker by training. Owns the twin's physics, the grade models and the translation between the mill and the platform.
Writing
Recent from the blog
What we publish when we learn something worth publishing.
Why break prediction needs 90 seconds, not 9
Lead time is the whole product. A model that detects a break as it happens is a very expensive event log.
The steam schedule should move before the basis weight
Why the standard grade-change sequence creates an avoidable moisture excursion, and what the twin found instead.
Nobody trusts a model. Everybody trusts a timestamp.
What actually changed crew attitudes during our first supervised-write deployment.
From the control room
What machine tenders say
Autonomy earns trust one shift at a time. These are design-partner quotes from pilot deployments [ASPIRATIONAL].
"The first thing that convinced the crew wasn't the control — it was the log. You can scroll back and see exactly why it dropped the nip. Nobody argues with a timestamp."
Machine tender · PM4 · Nordkraft Mills
"We had two people who could do a clean 135-to-110 transition. One retired in March. The twin now does the sequencing and the second one supervises it."
Production manager · Aurora Board
"Break prediction was the wedge. Ninety seconds of warning is the difference between a nip trim and four hours of threading."
Process engineer · Ternvik Paper
Guardrails
An agent that can move a nip needs a leash
Pulpum writes to production equipment. Every capability is scoped, every write is policy-checked, and every action is written to an append-only audit log the mill owns.
- Bounded action space. Each agent can only write to an explicit tag allow-list, inside per-tag rate and magnitude limits.
- Policy engine before every write. Autonomy level, shift, grade, interlock state and operator presence are all evaluated before a setpoint moves.
- Human-in-the-loop gates. Anything above the site threshold — speed ramps, grade releases, safety-adjacent moves — waits for a named approver.
- Immutable audit log. Append-only, hash-chained, exportable, and retained on the mill's own storage.
- Hard fallback. Loss of the edge node, the network or the model returns control to the DCS's last known-good state within one scan cycle.
- Tenant and IP isolation. Furnish recipes, grade models and defect libraries never cross a customer boundary. On-prem deployment available.
Compliance posture
| Standard | Scope | Status |
|---|---|---|
| SOC 2 Type II | Cloud control plane | RUNNING In progress [ASPIRATIONAL] |
| ISO 27001 | Company-wide ISMS | QUEUED Planned [ASPIRATIONAL] |
| IEC 62443 | Mill-edge OT security | RUNNING Design-aligned |
| GDPR | Operator data | SUCCEEDED Compliant |
| ISO 9001 / FSC | Quality + chain of custody records | SUCCEEDED Supported |
FAQ
Straight answers
The questions mill managers and process engineers actually ask in the first meeting.
Yes, but only within an explicit tag allow-list with per-tag rate and magnitude limits, and only at the autonomy level your site has set. Level 1 is advisory-only: Pulpum recommends and a human enters everything. Most mills spend their first weeks there before enabling supervised writes.
Control returns to the DCS last known-good state within one scan cycle. Pulpum is designed as a supervisory layer on top of your existing control system, never as a replacement for it, so a Pulpum outage degrades the mill to its current way of running — not to a stop.
Break prediction and defect classification typically need 8 to 12 weeks of QCS, DCS and inspection history per grade family, plus labelled break events. Advisory recommendations start in week one from the physics-based twin, and improve as mill-specific history accumulates.
Only if you choose cloud training. Recipes, grade models and defect libraries are tenant-isolated and never used to train another customer's models. A fully on-prem deployment with an air-gapped mill edge is available for sensitive producers.
You are, the same as with any control strategy — which is why every write is policy-checked, bounded, logged and reversible, and why anything above your risk threshold waits for a named approver. The audit log records the request, the reasoning, the limits applied and the human decision.
A 90 to 120 day mill-edge deployment on one paper machine, scoped to a single workflow with a pre-agreed baseline [ASPIRATIONAL]. Weeks 1–3 are connection and shadow-mode observation; weeks 4–8 advisory; weeks 9+ supervised or bounded writes if the mill is satisfied with the recommendations.
Get started
Come build the mill's operating system
We are hiring perception, control and papermaking engineers, and taking a small number of additional design-partner mills.