How do we combine our experts' knowledge with AI — and how can you?
Head of Data & AI · Lechler GmbH
A Swabian engineer and philanthropist. In 1879 he founded the trading house in Stuttgart that would grow into today's Lechler GmbH — and, in a sibling branch, ElringKlinger AG. The Lechler-Stiftung still funds his charitable work today.
A trading house for technical products, lubricating oils, and sealings. Patents for liquid atomization arrive in 1893. Own nozzle production begins in 1915 in Stuttgart-Berg, Bad Cannstatt, and Stammheim.
Europe's largest nozzle manufacturer. The spray that lands on metal sheets, vineyards, chemical reactors — and a thousand other places nobody thinks about.
Every droplet counts
Crop protection with minimal drift — boom nozzles that put the active agent on the leaf, not into the air.
Chemistry at plant scale
Gas cooling, scrubbing and flue-gas desulfurisation — spray systems engineered into the heart of the process.
Precision does the dirty work
Tank and surface cleaning, coating, cooling — high-impact jets that strip stubborn residue in seconds.
Built for liquid steel
Continuous casting, descaling and strip cooling — sprays that hold their geometry next to 1,200 °C.
Wherever a droplet has a job to do, a Lechler nozzle is doing it — from the open field to liquid steel.
Every sector speaks its own language. Domain-first development starts exactly there — with the people who speak it.
The Excel monster is just one example of experts helping themselves. The serious tools were built externally, professionally — but that road is expensive, and slow.
Spreadsheets, macros, little databases — deep domain knowledge on fragile foundations.
Works today, unmaintained tomorrow — and IT never sees it.
External engineering delivers quality — but not every tool justifies the invoice.
Even a small change means a new offer, new budget, and a place in someone else's queue.
Shrink the distance between an expert's need and a working tool — from months and budget rounds to days.
Realistic distribution in a typical mid-market company.
The expertise that defines a Lechler product doesn't live in a chatbot or a vendor's API. It lives in people — in their heads and on their screens.
The goal: put real tools into exactly these hands — without months of waiting.
No AI saves your company without the people on the shop floor — the ones who turn theory into a working installation, who know the customer site, who make the spec into a fact.
A trial balloon, not a rollout: a handful of domain experts, real tools, and a fair chance.
For a handful of domain experts — ask, draft, explore. No company-wide rollout, just a fair trial.
Frontend to API, wired together and ready — to build their own solutions instead of waiting.
Quasar · Vue · FastAPI · Python
Conventions, guardrails and templates — so what they build stays maintainable and safe.
The honest result: it didn't succeed. At all.
It knows nearly everything — the public theory, the textbooks. But not your nozzles, your bench data, or the findings your experts made beyond the literature. The output looks confident and technical. It is wrong in every specific.
The model retrieves real documents — the catalogue, technical references, bench measurements. The droplet-size answer is closer to right, but it still hallucinates the reasoning behind it.
Each one-shot failure produces a generalised rule that travels to the next task — not a hard-coded answer for the task we just lost.
Each distilled principle lands in a growing knowledge base — searchable by keyword and by meaning, and extended automatically with every round.
An AI that builds tools must test them itself — instantly. We built the stage for it: the AI runs smoke tests, reads the live UI and DOM, and drives the app like a user.
“You can’t optimize what you can’t measure.”
…and the faster you can measure, the more efficiently you can optimize.
Open source: github.com/Alyxion — llming-com · llming-stage
Machine engineering doesn't tolerate "mostly right". 98 % correct on droplet size is a catastrophic spray pattern in production.
Every time the LLM rewrites a tool, we replay every known answer. If any number drifts, we see it.
RAG, the knowledge DB, and a growing set of verified samples feed one loop — every failure makes the next attempt stronger.
Everything ships one-shot — or it doesn't ship.
AI-assisted CAD — from natural-language brief to validated 3D geometry
AI Build Steps
Specifications
Uniformity Score
—
Thread Size
Spray Angle
Descaling Setup
Benefits
This is where it all lands: the Lechler online-tools portal.
Core tools — owned and curated by the experts who know them best.
Droplets — tools built with AI by domain experts, open to every Lechler engineer.
From working prototype to published tool: one click, not one project.
The expert has the knowledge. Now they have the platform.
Precision nozzles for agriculture, industry, process technologies and metallurgy — 147 years of spray expertise, on every continent.
I help teams build, deploy and scale their own tool platforms and agentic AI solutions — from the first trial balloon to verified production.