AI Tooling
for the
Casting Floor
Ondrik works with foundries and precision machining plants in the northern corridor. We build practical AI tools — defect image review, production data structuring, and staff technology programmes — that fit into how your floor already runs.
Start a Conversation
Built for Plants That Make Things
We started Ondrik after spending time inside production facilities across Bukit Minyak and Batu Kawan — watching experienced inspectors flag surface cracks on cast parts by eye, production engineers pulling scrap counts from three separate spreadsheets, and supervisors trying to explain new tooling to staff across different shifts in two languages. The gap between what was possible with modern AI and what plants were actually using was wider than it should be.
Ondrik fills that gap. We scope what your floor needs, build tooling that fits your existing systems and workflows, and stay involved long enough that your team is genuinely comfortable using it. No steep onboarding, no black-box outputs the inspector can't challenge, no dashboards your production engineer has to chase a vendor to update.
Three Production Challenges, One Focused Supplier
Surface Defect Detection
Photograph a cast part, upload it, get a flag. Our image review routine points inspectors toward likely defect zones so they spend time confirming rather than hunting. The final call stays with the inspector.
Learn more →
Production Data Clarity
Machine logs, job cards, scrap tallies and inspection results consolidated into one structured store. Your production engineer can query yield and cycle time by part family without chasing three different systems.
Learn more →
Technology Adoption
New tooling only works when the people using it understand it. Our shop floor programme brings supervisors, inspectors and engineers into a common way of reading flags, logging findings and running daily reviews.
Learn more →AI Tools That Earn Their Place on the Floor
First Review, Not Final Word
Every flag the model raises goes to a human inspector. We design the tools so your experienced staff retain authority over accept and reject decisions.
Works with What You Have
We connect to existing machine logs, ERP exports and paper-based records. You don't need to replace your current systems to get structured, queryable data.
Delivery on a Production Calendar
Four-week image review, eight-week data structuring, ten-week adoption programme. Milestones are sized to fit around shift schedules, not against them.
A Model Trained on Your Part Families
Generic defect detection doesn't translate well to specific alloys and part geometries. We start by reviewing a sample of your own photographs — good parts, rejected parts, borderline calls — and train the first model iteration on those examples. The output is a review screen your inspectors can use from the first day of handover.
Each accept and reject logged through the screen becomes part of an accuracy report. If the model's miss rate on a particular defect type is higher than expected, the report says so clearly and points to a scale-up path.
Discuss Your Setup
Five Source Systems, One Query Layer
Most plants in the Simpang Ampat area have records in at least three places: a machine log the operator fills on a terminal, a job card on paper that gets filed at shift end, and a scrap tally in a spreadsheet that someone emails to the supervisor weekly. None of them talk to each other.
We inventory what you have, build connectors for up to five source systems, set validation rules and hand over a query layer the production engineer can use to pull yield trends by part family or cycle time by machine ID — without needing a data analyst on call.
Talk to the TeamWhat to Expect When You Work with Ondrik
We work with a small number of plants at a time. Every engagement starts with a scoping call, a site visit if practical, and a written scope document before any work begins.
Human-in-the-Loop Design
Every tool we build keeps the decision with your inspector or engineer. The model supports; it doesn't replace.
Written Scopes and Fixed Durations
Four-week, eight-week, ten-week — each programme has a defined scope document and a clear endpoint.
Bilingual Station Materials
Station cards and session materials are prepared in both Bahasa Malaysia and English so nothing is lost across your shift teams.
Accuracy Reports Included
Every image review engagement closes with a written accuracy report. You know how the model performed before you decide on next steps.
How We Think About This Work
Specificity over Speed
We scope carefully before we build. A model trained on the wrong sample set wastes everyone's time. Getting the scope document right in week one is worth more than a fast first demo.
Staff Confidence as a Deliverable
A tool nobody uses is a tool that failed. Our adoption programme isn't an add-on — it's built into every engagement that involves new tooling on the floor.
Data You Can Query, Not Just Read
Every structured data store we build comes with a query layer sized for production engineers — not data analysts. If the production engineer can't pull a yield report on their own at the end of the engagement, we haven't finished.
Three Programmes, One Foundry Supplier
Each programme is scoped, delivered and closed within a defined period. Pricing is fixed per engagement.
Casting Defect Image Review
An image review routine that examines photographs of cast and machined parts and flags likely surface defects for a human inspector to confirm. Covers sample capture protocol, lighting guidance, a first model trained on client examples, and a review screen with accept and reject logging.
- Accuracy report included
- Scale-up note on completion
- 4-week delivery
Production Data Structuring
Consolidation of machine logs, job cards, scrap records and inspection results into one structured store. Covers source inventory, connector development for up to five systems, unified schema, validation rules and a query layer for the production engineer.
- Data dictionary on handover
- Two training sessions
- 8-week delivery
Shop Floor Technology Programme
Twelve sessions across shifts, station cards in Bahasa Malaysia and English, a daily meeting format, coaching for three internal leads, and a review visit at week ten. Designed for plants of 50–400 staff.
- Written adoption report
- Year-ahead planning note
- 10-week structured delivery
The Pour Sequence
Scoping Call & Site Review
A 45-minute conversation to understand your part families, current inspection workflow and data landscape. We follow up with a written scope document — no commitment required at this stage.
Build & Integration
We build the tooling to the agreed scope, connect to your existing systems and run accuracy checks before handover. Weekly check-in notes keep your production manager informed.
Handover & Adoption
Handover includes a written report and at least two training sessions. For adoption-track engagements, we stay on-site through week ten and conduct a review visit before closing the scope.
From the Plants We've Worked With
"The image review screen took about three days for the inspectors to get comfortable with. After that first week they were flagging disagreements with the model and sending them back to Ondrik — which is exactly what we wanted. We're not running blind anymore."
Rozaimi bin Zakaria
QC Manager, Bukit Minyak Foundry
July 2025
"We had records in five places — two Excel files, an old DOS-era terminal export, paper job cards and an ERP module nobody really trusted. Eight weeks later there's one place I go to pull part-family yield. That's the difference."
Lim Kah Wai
Production Engineer, Simpang Ampat
14 July 2025
"The station cards in Bahasa Malaysia made a real difference on the evening shift. The supervisors on that shift were the most skeptical before we started and by week six they were the ones pushing for the week-ten review visit."
Norhafizah Fauzi
Plant Manager, Batu Kawan
22 July 2025
FAQ
Do you need to install software on our production machines?
How many part photographs do you need to train the first model?
Our records are partly on paper. Can you still structure the data?
What happens after the engagement closes?
How is our production data handled and stored?
Start the Conversation
Ondrik
14100 Simpang Ampat
Pulau Pinang, Malaysia
Office Hours
Monday – Friday: 8:30 am – 6:00 pm
Saturday: 9:00 am – 1:00 pm
Sunday & Public Holidays: Closed