Your best people's knowledge lives in their heads and their spreadsheets. We turn it into an assistant your operators actually use.

We understand the workflow, connect your documents and systems, prove the assistant is useful, and keep it reliable as things change.

Built and deployed inside a Fortune 500 manufacturer. Now for plants of 100 to 2,000 people.

Book a 30-minute call.

Real operator-assistant demo showing a question about draining an air compressor, an answer with safety prerequisites, and supporting passages from manual page 12. — A real answer from our operator-assistant demo, alongside the manual passages it cites. Public equipment manual; no customer documents. View full-size screenshot (opens in a new tab).

Original answer panel, including the power-off and unplugging prerequisites, drain-valve warning, four cited steps and daily drainage guidance. — A real answer from our operator-assistant demo, alongside the manual passages it cites. Public equipment manual; no customer documents. View full-size screenshot (opens in a new tab).

97% accuracy across seven product lines
51% of support issues resolved without a ticket
A global AI team of 15+ across six locations

Historical results from a Fortune 500 manufacturer. See the proof.

How we make AI work for your team.

Understand the work.
Which questions matter, who needs the answers, and who approves them.
Connect the pieces.
Your documents, permissions and existing systems working together.
Prove it helps.
Acceptance tests and measures for adoption, resolution time or escalations.
Keep it dependable.
Current sources, clear ownership and retesting after changes.

We agree the measures, document owners and retesting responsibilities before handover. Ongoing support is scoped and priced separately.

The problem

When the person who knows isn't there.

Your experts are retiring, taking thirty years of knowledge with them.
Your procedures are buried in PDFs, Excel and SharePoint. Answers take a day, sometimes two weeks.
Your new hires keep asking the same person the same questions.

Real operator-assistant demo asked for a Toyota forklift tire pressure. It responds: I don't know. That isn't in the manuals I have. No source passages are cited. — When the manuals do not cover the question, the assistant says so. In a deployed build, it then passes the question to a named expert or a specialist queue. A real response from our demo, using public equipment manuals. View full-size screenshot (opens in a new tab).

Answer: I don't know. That isn't in the manuals I have. — When the manuals do not cover the question, the assistant says so. In a deployed build, it then passes the question to a named expert or a specialist queue. A real response from our demo, using public equipment manuals. View full-size screenshot (opens in a new tab).

How it works

One department. One working assistant.

One department. One body of documents. A working assistant in 6 to 8 weeks. Fixed price. You own it.

Builds start at $45,000 or €40,000.

Start with a ten-day pilot.

Or train your team first with a one-day testing workshop.

We agree the scope, price and what a successful handover looks like before work starts. The build begins once the agreed documents and IT access are in place.

What the build includes

We also agree which knowledge needs to be written down from your experienced people. Your team checks it before it becomes a source for the assistant.

Your existing documents
Procedures, spreadsheets, scanned PDFs, photos and diagrams with handwritten notes. Pages that cannot be read reliably are flagged for review.
Answers that show their source
Your operators can check the document, page or sheet behind an answer.
A specialist when the documents run out
When the documents do not support an answer, the assistant says so and passes the question to a named expert or a specialist queue on your team. Both steps are tested before handover.
IT and security review
We agree where the documents go, who can access them and how the assistant fits your systems.
Data architecture agreed before the build
Where documents are stored, how they are indexed and how updates flow are settled in the first week, so the build does not change direction midway.
Testing with your operators
The people doing the work try it on real questions before handover.
A handover your people can use
You receive the project code, setup notes and instructions for updating documents and keeping it running.

What you own

Your knowledge stays yours.

Every assistant has three layers. We build the first two as assets your plant owns.

Your knowledge. Procedures, manuals, spreadsheets and the rules your experienced people carry, written down and approved by your team.
Your workflows. The questions that matter, who approves the answers, and the acceptance tests that show the assistant is working.
The AI model. Chosen to fit your IT and security requirements. Your approved sources, expert rules and tests are handed over with the project.

Software licences, hosting and any ongoing support are agreed separately before work starts. The assistant helps people find approved information; it does not replace your plant's safety procedures or required checks.

The first version of this assistant went from idea to a working pilot on the line in three days: two to build, one to deploy. Our earlier internal AI development cycle had taken 18 to 24 months. That initial pilot is distinct from the scoped six-to-eight-week client build offered here.

Ways to start

Choose the first step that fits.

Workshop

Format: One day, on site or private online, up to ten people
Price: $3,500 or €3,000 per team
Credit: €1,000 off an Assistant Pilot started within 30 days
Best for: Teams learning to test AI before they trust it

See the workshop

Assistant Pilot

Format: Ten days, one document set, up to 300 pages, five users, 50 test questions, thirty days of access
Price: $4,900 or €4,500, fixed
Credit: Credited in full toward a build signed within 30 days
Best for: Proving one document set with real questions

See the Assistant Pilot

Department build

Format: One department, one body of documents, 6 to 8 weeks
Price: Builds start at $45,000 or €40,000
Credit: Fixed price, you own it
Best for: A working assistant your operators use

Book a 30-minute call

Proof

Built. Tested. Put to work.

Results from my previous role leading AI inside a Fortune 500 manufacturer. These are historical results, not promised results for your plant.

Product knowledge
Turned 4,000+ technical documents into an assistant that answers engineers' product questions in seconds instead of days: 97% accuracy across seven product lines, 100% on critical equipment, tested on about 2,000 real questions.
Work instructions
At a plant where 15% of line stoppages traced back to procedure errors, built an assistant that reads the actual work instructions, text, tables and photos, and answers operators on the line in two languages.
Troubleshooting
Launched a customer self-service troubleshooting tool that resolved 51% of issues without a ticket and cut technical-support chat volume by about half in its first six weeks.

Background

Where my experience comes from

Emerson Automation Solutions, measurement division. Built and led a global AI/ML team of 15+ AI developers, data scientists, and AI engineers across six locations, set up AI governance and a stage-gated path to production, and ran an innovation lab that cut AI development cycles from 18–24 months to 2–4 weeks. This is where I built the operator knowledge assistant this practice is based on.

Carl Zeiss AG, AI Program Manager. Led machine-vision defect detection deployed at a major contract manufacturer in China, inspecting cover glass for a flagship smartphone. The system caught scratches and other surface defects on the line, at the volume and precision that product demands. In industrial 3D printing, led a program that built production parts, such as pistons for a German luxury sports car maker and orthopedic knee implants, and delivered software to test the quality of those parts.

Earlier: Program management at Medtronic in regulated medical devices, and ASIC engineering at IBM.

Cornell University. Certificates in product management, program leadership and executive leadership.

Florida International University. Electrical Engineering and Engineering Management.

Columbia University. Supply Chain Management.

Consumer electronics, automotive, medical devices, industrial instrumentation: every one of these was built where the work happens, on production lines with real operators and real quality targets.

First page of the public ifm IM0010 datasheet, showing a 20 mm sensing range and a 40 × 40 × 120 mm rectangular sensor. — 20 mm sensing range. An actual value in an actual document. Public ifm IM0010 datasheet, dated 6 March 2003, in French, published by Audin. Shown as a document example, not an assistant answer or operating instruction.

Why not do it yourself

You could start with a procedure and ChatGPT.

You could paste a procedure into ChatGPT today. For one person and one document, that may be enough.

The harder part is the plumbing: IT approval, access to the right files, scanned pages and handwritten notes. It also means checking that the assistant has not invented a torque spec, and keeping it running when the person who set it up gets busy.

You do not need to build a permanent AI team for one defined project. This is the AI team you can't hire, for the one project you need.

Who's behind this

Virginia Kalesić

I led AI for a Fortune 500 manufacturer's measurement division, reporting to VPs and general managers. Earlier, I worked in program management at Medtronic and Zeiss, and in engineering at IBM.

I built and led a global team of more than 15 people across six sites. I set up the AI Governance Board and an approval process with checks at each stage, so projects were reviewed before they went into use.

I also built an innovation lab that took development cycles from 18 to 24 months down to 2 to 4 weeks, with working prototypes people could test.

Also in my portfolio

Oversaw a maintenance troubleshooting assistant for technicians at a manufacturing site, built on the plant's own repair history and expert input, with translation support for Chinese, English, and 16 other languages, live since April 2025 and running for about $200 a month plus token costs.

Oversaw an operations platform, live since August 2025, saving about $160K a year and rolled out to every site in the division.

I lead every engagement personally, working with our permanent team across systems architecture, AI engineering and testing.

I work from Europe, across CET and PST.

Founder-led delivery

The team behind the work.

I lead every engagement personally, supported by a dedicated engineering team. Our responsibilities cover systems architecture, AI implementation, evaluation and ongoing reliability.

Systems architecture

Technical design, integrations, access controls, deployment planning and production-readiness reviews.

Senior AI engineering

Document processing, retrieval, citations, evaluation tests and improvements as sources or requirements change.

Before we start

A clearly scoped build.

What does the fixed price cover?

We agree the department, source documents, intended users, delivery milestones and acceptance tests before work begins. Additional departments or integrations need a separate scope. Builds start at $45,000 or €40,000, invoiced in your currency at the listed price.

How do we handle access and sensitive information?

We review document access, hosting and security requirements with your team before selecting the implementation. Do not send confidential manuals through a public inquiry form.

What happens at handover?

The proposal specifies ownership, documentation, operating responsibilities and any ongoing support. Your team should know who maintains the sources and who approves changes.

Can we see it work first?

Arrange a guided operator-assistant demonstration to see source citations and how the assistant handles questions its manual cannot answer.

What does your plant need to know when your best person is out?

Start with one department and the questions that keep coming up. We'll talk through the documents you have and whether a fixed-price build is a good fit.