Teh Tarik Digital · Sungai Petani, Kedah · Kulim Industrial Corridor

AI That Covers the Line You Can't Staff

Kulim Hi-Tech Park is doubling to 12,000 acres. Infineon's world's-largest 200mm silicon carbide power fab here — a RM30.1 billion investment — has been in production since 2024. Ferrotec's second Kulim plant broke ground in April 2025. The corridor was scaling before any of this made headlines, and it's still outpacing the workforce meant to run it.

At SEMICON Southeast Asia 2026, the industry named the real bottleneck: a tenfold gap between engineer supply and demand, 12,000–15,000 additional technical roles needed by year-end, current pipelines meeting barely 60–65% of it. On 1 June 2026, Employment Pass salary floors rose across the board — manufacturing technical roles now carry a RM 7,000 floor. Headcount for QC just got harder to find and more expensive to keep. We're deployed on-site in Sungai Petani, inside the corridor, not visiting it — and we build the systems that let plants stop competing for engineers who don't exist.

12K+
acres, Kulim Hi-Tech Park expansion
10x
engineer supply-demand gap
RM7,000
new EP technical wage floor
60–65%
of demand current pipelines meet

The Corridor Was Already Scaling

// SIGNALS

Kedah doesn't get the AI-conference coverage Penang does — this isn't AI-scene news, it's industrial-investment news, and that's exactly why the labor-gap story above is real and not a talking point.

Infineon's RM30.1 billion 200mm silicon carbide power fab at Kulim Hi-Tech Park — the world's largest of its kind — has been in production since 2024, and Ferrotec's second Kulim plant broke ground in April 2025 with a roughly year-long build timeline. Both are the clearest signal of how fast this corridor was already scaling before the labor conversation caught up to it. KHTP2's land release (Zone 4A, 247 acres across 10 lots) is still working through allocation. Sources: Infineon · MIDA · Kulim Hi-Tech Park announcements

What We Build Here

// CAPABILITIES

We focus on the Alor Setar–Sungai Petani stretch of the corridor, closest to Kulim Hi-Tech Park itself — not a generic Malaysia-wide pitch, a build scoped to what's actually running here. Four capabilities, and none of them are speculative: computer vision QC is the direct answer to a line that can't find inspectors; shift and yield reporting exists because someone on your floor is still compiling that by hand at the end of every shift; predictive maintenance and plant workflow automation extend the same discipline to the facility infrastructure and the paperwork around it, not just the production line itself.

01

Computer Vision QC

Defect detection trained on your actual production line — not a generic pretrained model that misses the specific flaws your line produces. This is the direct answer to a floor that can't hire enough inspectors to hold a consistent standard shift after shift.

02

Shift & Yield Reporting

Auto-generated from CV output the moment a shift ends, not compiled by hand at 11pm by whoever's still on the floor. The report exists because the inspection already happened — it's a byproduct of the system working, not a separate task someone has to remember to do.

03

Predictive Maintenance

SmartPole OS-style health scoring, where facility infrastructure is in scope — the same 85%+ fault-prediction approach that already runs on live poles, applied to plant equipment so a failure gets flagged days before it becomes downtime.

04

Plant Workflow Automation

Approvals, maintenance dispatch, and compliance logging — the paperwork layer around the production line that usually gets deprioritised until an audit forces someone to reconstruct it from memory.

Why Teh Tarik Digital, not a vision-hardware vendor: point-solution vision systems solve one line. We integrate into what you already run — existing cameras, existing MES/ERP, existing reporting — so you're not adding another vendor portal to your stack.

Benefits & Results

// PROOF

The labor shortage isn't a future risk here — it's already priced into your headcount budget. The question isn't whether automation helps; it's whether the numbers you're being shown are real or a projection dressed up as a result. The left column is what actually changes for the people on the floor. The right column is where those changes have already been measured on live products — SmartPole OS's fault-prediction numbers, IRIS's audit logs — not modeled for a pitch deck.

What Changes

  • Covers the line you can't staffConsistent inspection regardless of the local engineer shortage.
  • No headcount scramble at RM 7,000+ wage floorsAutomation absorbs QC demand without proportional hiring cost.
  • Fewer missed defectsNo fatigue, no shift-to-shift inconsistency.
  • Faster shift reportingAuto-generated, not manually compiled.
  • No rip-and-replaceIntegrates into existing cameras, MES, ERP.

What We've Measured

  • SmartPole OS — 85%+ / 40%85%+ fault prediction accuracy on a 7-day horizon via ML health scoring, and a 40% reduction in incident response time — the same predictive-maintenance approach applies to plant-floor infrastructure.
  • Dual-path failoverHigh-availability architecture with failover under 30 seconds; exact uptime SLA is defined per deployment, not a fixed marketing number.
  • IRIS — audit-readyLive orchestration platform with immutable, exportable audit logs on every action.

How It Works

// PROCESS

Four stages, starting with your actual line, not a generic deployment checklist. The order matters: we won't propose a CV model before we've audited your defect taxonomy, and we won't integrate into MES/ERP before the model's already trained on your real defect classes. Skipping ahead is how vendors end up selling you a system that works in the demo and breaks on your actual production floor.

01

Assess

Line audit, defect taxonomy, existing camera and hardware inventory — we're not proposing anything until we've seen what's actually running on your floor.

02

Build

CV model trained on your actual defect classes, not generic datasets — coverage that matches the flaws your line actually produces, not a stock library of defect types that mostly don't apply.

03

Integrate

Connects to line hardware and MES/ERP without replacing what's already there — the deployment risk is in the integration, not the model, so this is where most of the real work happens.

04

Operate

Live QC dashboard, auto shift reports, escalation on flagged defects — and the same team that scoped the project stays on for support, not a handoff to a different account manager.

// FROM THE LAB

Real Builds, Real Numbers

How we actually build multi-agent systems — the architecture decisions, not the highlight reel.

MULTI-AGENT SYSTEM

6 Months of Iteration, One Working Multi-Agent System

The supervisor-specialist rebuild behind IRIS v2 — the state-schema-first approach that replaced the original single-agent design. Full case study →

FAQ

// 10 QUESTIONS

Straight answers for plant managers evaluating whether this actually fits a line that's already running lean.

Why computer vision instead of hiring more QC staff?
The engineer supply gap in this corridor is real and dated — 12,000–15,000 unfilled technical roles by end of 2026, with pipelines meeting barely 60–65% of demand, and wage floors that just rose. Computer vision covers the line you can't staff. See our Lab breakdown on manufacturing computer vision.
Do you replace our existing camera/line hardware?
No — we integrate into what's already installed rather than requiring new hardware.
How long does a line audit and deployment take?
Depends on line complexity and defect variety; the line audit itself is the first step and scopes the actual timeline.
What actually happens to the defect data after the camera catches it?
It's not just a photo and an alert — flagged defects feed into shift and yield reports automatically, coordinated by the same kind of multi-agent handoff that routes maintenance dispatch and compliance logging. See our multi-agent orchestration breakdown.
Can this integrate with our existing MES/ERP without a middleware project?
Yes — integration is scoped during the assess phase against whatever you're already running. See our breakdown of MCP (Model Context Protocol) for the technical detail.
Is this the same as the chatbots and copilots we keep hearing about?
No — a line QC system executes: it inspects, flags, and logs without waiting on a conversation. See our model comparison piece for the distinction.
Does this only work for semiconductor lines, or other manufacturing too?
The core computer vision and workflow automation approach applies to any production line with a defined defect taxonomy — we've applied the same architecture to logistics and warehousing. See our logistics automation breakdown.
Do you offer anything beyond the production floor — dispatch, compliance logging, approvals?
Yes — workflow automation extends to maintenance dispatch, approvals, and compliance logging, coordinated by multiple specialised agents. See our multi-agent build walkthrough.
Is this PDPA-compliant for a facility handling worker or visitor data?
Yes — where any people-facing monitoring is in scope, it's built PDPA-aligned by design. Full detail: Semiconductor & Electronics AI.
Who is Teh Tarik Digital?
Founded in 2016 by engineers with backgrounds spanning international cryptocurrency exchange operations, BNM-adjacent banking infrastructure, and government technology initiatives at the ministerial level.

Tell us what's bottlenecking on your line — we'll tell you what it takes to cover it.