Basel / Zürich · AI product builder for manufacturing operations

I build the AI agents behind industrial operations.

Senior Process Engineer · Project Manager · Hitachi Energy PhD with distinction · AGH · Cambridge Judge 2025 AI agents, built and in daily use

Ten years across research, product development and manufacturing. A 10 MCHF portfolio, scrap from 15% to 1% on a running line, a world-first 420 kV SF₆-free breaker. I also build the AI agents that run my own project and KPI reporting. AI people rarely know the factory, and factory people rarely build agents. I do both, and I know where it gets hard: a shift that trusts millimetre paper has to come to trust a screen.

See the trajectory → What I build → LinkedIn ↗
Portrait of Dr. Paweł Krawczyk
By the numbers

Specifics, in the absence of which an entrepreneur is just a hopeful adjective.

Portfolio I lead today
MCHF
Four CAPEX, OPEX and R&D projects at Hitachi Energy, from business case to commissioning.
4 projects · Steering Committee and CXO reporting
Scrap on a running line
1.2MCHF/yr
Saved each year on a ceramics line. Glazing yield to 96% on the same line.
DMAIC · SPC · Python on the line data
Team I lead
30 people: 19 internal, 11 external, across engineering, R&D, supply chain, automation, IT/OT and sales.
World first
kV
SF6-free breaker (EconiQ™). I solved its launch-blocking gas issue in 4 weeks.
Suppliers and sites
CHSEPLCNUSIT
Six countries in one portfolio, from first drawing to commissioning.
AI agents in daily use
My project and KPI reporting runs on agents and Python scripts I built. The log above is illustrative.
Languages at work
  • Polish
  • English
  • German
  • Russian
Research
+citations
5 peer-reviewed papers, 3 patents, h-index 4.
Now ·

What I am working on this quarter.

  • Leadinga capacity expansion toward 1.5 M parts a year at Hitachi Energy, Wettingen.
  • BuildingAI agents that automate project and KPI reporting for a 10 MCHF portfolio.
  • Growing5 junior engineers and interns, with the qualification criteria and training material I wrote.
  • Studyingfor Lean Six Sigma Green Belt and PMP, both in progress.
How I build AI for operations

Four steps, in the order a factory can live with.

01

Find the constraint

Every line has one step that sets the pace for everything after it. I find it first, with the Five Focusing Steps and the line data, before anyone buys software.

02

Build the smallest agent that watches it

One agent, one job: read the data around the constraint and say what changed. Built in Python and Claude Code, tested on the real line, not on a slide.

03

Put it in front of the people who own the line

A shift that trusts millimetre paper has to come to trust a screen. That happens on the floor, with the people who will use it, one catch at a time.

04

Measure it in the factory's own units

Scrap, yield, hours, lead time. If the number does not move, the agent goes. On my own line, scrap went from 15% to 1%.

Interactive model. Click a station to give it a second head and watch the constraint move. Cycle times are illustrative.

Experience

A research bench. Then a factory. Then ten million francs of projects across six countries. One instrument, scaled.

Read top down: newest first. Each role was a larger version of the previous problem. Find the constraint, instrument it, ship the change.

11/2024 to present · Wettingen, CH

Senior Process Engineer · Project Manager

Hitachi Energy · Surge arresters, 3× capacity expansion toward 1.5 M parts/yr
  • Own business case, scope, schedule and budget of four CAPEX, OPEX and R&D projects worth 10 MCHF.
  • Lead a project team of 30 (19 internal, 11 external). Suppliers in CH / SE / PL / CN / US / IT.
  • Scrap from 15% to 1% on a running ceramics line, worth 1.2 MCHF/yr. Glazing yield to 96%.
  • Built AI agents and Python scripts that automate project and KPI reporting.
  • Wrote the commissioning playbook (FAT/SAT, IQ/OQ/PQ, MES/OEE) now used across the division.
  • Cambridge Judge, General Management certificate, "Steering Complex Projects" (2025).
2025 to present · Switzerland / Poland

Co-founder & AI Architect · JustAutomate

JustAutomate · Side venture, declared to and approved by my employer
  • Design the agent systems and the method behind them, for manufacturing SMEs.
  • BOND: agentic layer that surfaces a factory's throughput constraint from live data.
  • Co-founder alongside Paweł Pasterak (CEO).
01/2024 to 07/2024 · Ludvika, SE

R&D Senior Engineer · Task-Force Project Leader

Hitachi Energy · Lead-free coating programme, team of 10
  • Recovered a 9-month delay in under 6 months. Next Stage-Gate passed on the first attempt.
  • Lead-free laser coating now in series production in CH and SE: 95% less material waste, 30% lower cost, yield up to 95%.
  • Built the business case and a global scale-up blueprint for SE, CH and CN.
09/2020 to 01/2024 · Baden-Dättwil, CH

Associate Scientist · Semiconductor Packaging & Applied Materials

Hitachi Energy Corporate Research · PhD in parallel, AGH, with distinction
  • Key contributor to the world-first 420 kV SF₆-free breaker (EconiQ™). Solved a launch-blocking gas decomposition issue in 4 weeks.
  • 40-year lifetime models: fill-pressure spec cut 12% at constant reliability.
  • Qualified three alternative suppliers in about 3 months after the 3M PFAS exit.
  • Built and ran an analytics lab (150 kCHF budget). Developed 3 junior scientists.
  • Patent co-inventor: US20240372340A1 / EP4388562A1 (zeolite adsorber). CIGRE D1-10755 (2024) and D1 (2022).
01/2020 to 07/2020 · Dübendorf

Master Project Researcher · High Performance Ceramics

Empa (ETH Domain) · MSc Chemical Technology, AGH
  • Nanocomposite filtration media for virus removal (6-log).
  • Two papers and a patent filing.
2017 to 12/2019 · Krakow

Warehouse and Inventory Team Lead

Vesuvius PLC · Refractory ceramics · BSc Materials Engineering, AGH
  • Led a first-line team of 10.
  • 500+ t/day material flow, OTIF over 99%, WIP down 35%.
  • DMAIC that cut a chronic scrap defect by 80%. ISO 9001:2015 internal auditor.
Expertise

Six domains. One operating principle. Find the constraint, then build the instrument that watches it.

Industrial projects & program leadership

10 MCHF today

CAPEX and OPEX, Stage-Gate, business cases, FAT/SAT, multi-site supplier coordination.

Theory of Constraints × operational diagnosis

in daily use

Five Focusing Steps run as a weekly cadence. Bottleneck-first thinking.

Materials science & chemistry

PhD · patents

Eco-gases, high-entropy materials, surface chemistry. The discipline of measuring before claiming.

Lean Six Sigma & change leadership

Green Belt in progress

DMAIC, FMEA, SPC, DOE. Yellow Belt held. Eight juniors and interns developed, a first-line team of 10 led.

AI product building for operations

in daily use

Agents for project and KPI reporting, Claude Code, Python. Built for the people who will use them, tested on the line.

Industrial data & systems

8 yrs

SAP ERP, MES/IoT, Power BI, MS Project. The plumbing the AI layer plugs into.

Live network. Drag a node, or hover a domain card to light its tools.

Case studies

Four problems, what I did, and what moved. Every number is on my CV.

Yield and quality · Wettingen15% → 1%

Scrap on a running ceramics line

Problem
A running ceramics line in a surge arrester factory was losing 15% of its parts to scrap.
What I did
Made the engineering calls on deviations and non-conformances: containment, root cause, permanent fix. Tracked OEE, yield, scrap and cycle time with my own Python analysis of the line data.
Result
Scrap at 1%, worth 1.2 MCHF a year. Glazing yield to 96% on the same line.
Task force · Ludvika9 → <6 mo

A delayed programme, recovered

Problem
A strategic development programme for a lead-free coating was 9 months late.
What I did
Came in as task-force lead with a team of 10. Re-baselined scope and schedule, built the business case and a global scale-up blueprint for SE, CH and CN.
Result
Delay recovered in under 6 months, next Stage-Gate passed on the first attempt. The laser coating now runs in series production in CH and SE: 95% less material waste, 30% lower cost.
High voltage · Baden4 weeks

A launch blocker on a world first

Problem
A gas decomposition issue was blocking the launch of EconiQ™, the world-first SF₆-free 420 kV circuit breaker.
What I did
Ran the analytics in a lab I had built and solved the issue in four weeks.
Result
Reported at CIGRE in 2022, later a granted patent family (US20240372340A1, EP4388562A1).
Supply chain · Corporate Research3 in ~3 mo

New suppliers after the 3M PFAS exit

Problem
When 3M left PFAS production, a qualified material disappeared from the supply chain.
What I did
Qualified three alternative suppliers in the EU, India and China.
Result
Done in about 3 months, followed by seminars and 6 global training webinars for 150+ participants.
Try it

The statistics behind scrap going from 15% to 1%.

A live control chart. Every dot is one part off the line; a part outside the tolerance band is scrap. Narrow the spread or move the centre and watch scrap and Cpk respond.

in tolerancescraptolerance limits
Expected scrap15.0%
Cpk0.48
Last 200 parts0

Illustrative process, real statistics: a centred normal process with tolerance at ±1. 15% scrap means σ ≈ 0.69 (Cpk 0.48); 1% means σ ≈ 0.39 (Cpk 0.86). On my line, the slider was root-cause work and a permanent fix.

Knowledge that stays

AI changes how a factory builds its know-how, finds it, and keeps it when people move on.

Most process knowledge sits in batch records, deviation reports, test results, shift notes and the heads of a few experienced people. I build semantic knowledge bases that turn all of it into one searchable memory. Every document becomes a point in a space of meaning, a question becomes a point too, and the answer is the closest one, whichever words were used.

When the process changes or a new study comes out, you add new points and the memory is current, without anyone redesigning a database. That is the part I specialise in: getting a factory's know-how out of folders and heads, into a memory an engineer or an agent can ask in plain language.

01

Search by meaning

"Cracks after firing in winter" finds the note about humidity during drying, even though they share no word.

02

Every answer cites its source

Each value carries a tag and the document it came from. A gap is reported as missing, never filled with a guess.

VERIFIEDCALCULATEDINDICATIVEASSUMPTIONMISSINGCONFLICTING
03

One place for every fact

Two copies of one fact and an agent cannot tell which is current. Each fact lives once, old copies go to an archive that can be restored.

Try it · demo knowledge base

Ask the factory a question.

Searches the documents and reads only the few pieces that match the question.

Keyword search
Semantic search · closest documents

    Demo with invented documents from an imaginary plant. Similarity here is computed over a small set of hand-tagged concepts, to show the idea; a real system uses a language model's embeddings over your own documents.

    AI agents a factory actually uses.

    Outside my role at Hitachi Energy I co-founded JustAutomate, where I design agent systems for manufacturing SMEs. BOND, the flagship, watches live operations data and names the biggest throughput constraint every Monday, with the evidence attached. Side venture, declared to and approved by my employer.

    "A shift that trusts millimetre paper starts trusting a screen the day the screen catches something they would have missed."
    Visit JustAutomate ↗ Talk about AI in operations →

    My own agent system as one machine, assembled in the order I built it. The press in the middle is the model; everything around it is the system. Click for the next step.

    Research footprint

    A small, dense body of peer-reviewed work. The kind a hiring director quietly checks.

    Flagship paper

    High-entropy oxide, first reported in its class

    The paper a materials reviewer recognises before reading the cover. Anchors work spanning high-entropy oxides, thin perovskite films and colloidal processing.

    First of its class
    Granted patent · US / EP

    Zeolite adsorber for eco-gas breakers

    Co-inventor on the adsorber chemistry that made the SF₆-free switching envelope hold under service conditions.

    US20240372340A1 · EP4388562A1
    Patents in preparation

    Bushing coating · low-temperature CO catalysis

    Two further filings underway alongside the EconiQ™ programme, both born out of lab work that started as a production problem.

    2 filings underway
    Peer-reviewed body

    5 papers, 110+ citations, h-index 4

    Materials · ACS Applied Electronic Materials · Membranes · Applied Surface Science. Small, dense, and verifiable.

    110+ citations · h-4
    CIGRE

    Conference reporting on SF₆-free switchgear

    Representative for the eco-gas programme at the industry body that sets the reference for high-voltage practice.

    D1-10755-2024 · D1 PS2/2.03
    Education

    PhD with distinction · Cambridge Judge

    AGH Krakow: BSc Materials Engineering 2019, MSc Chemical Technology 2020, doctorate in materials science 2024. Cambridge Judge, Steering Complex Projects, 2025.

    PhD with distinction
    Contact

    Bringing AI into manufacturing operations? Let us talk.

    I am glad to hear from people building AI into manufacturing and energy operations, whether that is a role, a project or a problem worth comparing notes on. I know the line, I build the agents myself, and I know how to get a shift to use them. Email reaches me directly, and I answer the question you asked.

    Direct line
    me@pawelkrawczyk.ch