Teh Tarik Digital · Penang CyberCity · Est. 2016

Built Where Malaysia's Tech Industry Actually Is

"AI transformation" pitches often come from a vendor who flies in, presents a deck, then leaves before anyone finds out whether the defect-detection model actually holds up on the line. We work differently because we're built into the same industrial base we're pitching to: our office sits in Penang CyberCity, and Intel, Infineon, Micron, Bosch, and AIXTRON's newest greenfield fab are neighbours, not case studies.

That proximity is the whole pitch. A Penang AI project succeeds or fails on details a remote vendor never sees — which line actually bottlenecks at 2am, which compliance officer needs to sign off before a camera goes live, which ERP integration always gets deprioritised until it breaks something. We're close enough to know, and we've been building here since 2016.

5%
of world semiconductor exports
250+
registered AI startups, Digital Penang
RM30.1B
Infineon Kulim SiC fab build-out
~45%
of Malaysia's national exports, Penang — Q2 2026

The State Is Moving. We're Already Inside It.

// SIGNALS

Penang isn't waiting for AI adoption to happen to it — the state is actively pushing it, and the push is showing up in real numbers, not just conference talk. We work across the island and the mainland: Bayan Lepas, Georgetown, Bayan Baru, Butterworth, and the wider Seberang Perai industrial belt, not just the district our office happens to sit in.

Penang's AI Digitalisation Conference (PDX2026) took place 15–16 July at Setia SPICE Convention Centre in Bayan Lepas — a short drive from our office — bringing government and industry together on the state's AI-adoption push. Digital Penang has also been running an AI Hackathon for Public Sector 2026, part of a broader push that's grown the state's registered startup base past 250, targeting 500 by 2030. Separately, Penang's Global Business Services sector made "AI in Action" its own 2025 conference theme — a sign of how fast AI-driven analytics is moving from pilot to expectation there. Sources: Digital Penang · Setia SPICE / PDX2026 event materials · Penang Institute

What We Build in Penang

// CAPABILITIES

Four verticals, one team — and none of them are here because they're trendy. Semiconductor and E&E are the obvious one; the sector's real problem right now isn't demand, it's a workforce that can't keep pace with it, so QC lines are running on fewer hands than they need. Enterprise and public-sector buyers have a different problem — procurement and approvals that move at the speed of the slowest signature in the chain. Knowledge work inside SMEs and larger teams has the opposite failure mode: too much institutional knowledge locked in one person's head, nowhere searchable. And medical devices, the largest sector of its kind in ASEAN here, has a compliance bar most AI vendors haven't actually cleared. Four different failure modes, four different builds — not one generic platform stretched four ways.

01 · SEMICONDUCTOR

Semiconductor & Electronics

Computer vision QC, defect detection, and floor automation for front-end and back-end E&E manufacturers — trained on your actual defect classes, not a generic dataset. Full breakdown →

02 · ENTERPRISE

Enterprise & Public Sector

Workflow orchestration for procurement, approvals, and audit-ready compliance — IRIS, live in production, built for BM+English documents and Malaysian data residency.

03 · KNOWLEDGE

Knowledge & Operations

Company-wide AI search and automation for SME and enterprise teams, self-hosted so your data never leaves your own VPC — JomWork.

04 · HEALTHCARE

Medical Devices & Healthcare

Penang's medical devices sector is the largest in ASEAN. Our clinical AI compliance work is the closest reference point — see NURI.

Benefits & Results

// PROOF

Every AI vendor shows up with a stat. Most of them don't survive the follow-up question of where the number actually came from. The left column below is what changes for the people doing the work — the inspector, the approver, the ops lead who gets paged at 2am. The right column is where those changes have already been measured, on live products, not a projection.

What Changes

  • Fewer missed defectsComputer vision holds a consistent inspection standard shift after shift — it doesn't get tired at hour 10, and it doesn't vary between your best inspector and a new hire.
  • No headcount scrambleWhen demand spikes, automation absorbs it. You're not suddenly hiring and training QC staff on a six-week runway.
  • Audit trail on every actionIRIS logs every approval, decision, and human intervention as an immutable, exportable record — the kind regulators and auditors actually want to see.
  • Human sign-off retainedIRIS surfaces information and flags risk; it doesn't make the final call. The person accountable stays the person deciding.
  • Scales without a rebuildConnects to existing ERP, procurement systems, SharePoint, and communication platforms via REST API or webhooks — you're not ripping anything out.

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 platform scales from a 50-pole campus to a 2,000-pole municipal network without a re-architecture.
  • NURI — 2.5+ hrs / 40%2.5+ hours of admin time reclaimed per clinician shift, and a 40% reduction in total documentation turnaround.
  • JomWork — self-hostedClient data never leaves their own VPC — the non-negotiable most enterprise buyers actually lead with.

How It Works

// PROCESS

Most AI vendors sell you a package before they've seen your floor. We do it backwards on purpose — assess first, because the right build for a 50-camera semiconductor line looks nothing like the right build for a procurement team drowning in PDF approvals. Four stages, same team throughout, no handoff to a different account manager once the contract's signed.

01

Assess

We start by figuring out which vertical actually fits — semiconductor, medical device, enterprise, or general SME — and audit what you're already running before proposing anything.

02

Design

We scope the right solution to that vertical: computer vision, workflow orchestration, or knowledge automation — not a one-size-fits-all package.

03

Build & Integrate

We connect to what you already run. No rip-and-replace, no forcing your team onto a new ERP just to use our tooling.

04

Operate

Live deployment with ongoing support — the same team that scoped the project is the one you call when something needs adjusting.

What to Look For in an AI Automation Partner

// DILIGENCE

Penang's AI market has gotten crowded fast — which is good for buyers, but it also means slowing down on a few questions before signing anything.

Can they name the actual product you'd be using, not just a category of service?
Is there one team accountable end-to-end — from the first conversation through ongoing support — or does the relationship change hands after the sale?
Has the company operated long enough, in regulated-enough environments, that its compliance claims are backed by real experience rather than marketing copy?

Teh Tarik Digital has answered those questions the same way since 2016: named products (IRIS, JomWork, SmartPole OS, NURI) that we built and support ourselves, one accountable team from scoping through deployment, and a founding background in regulated, cross-border, high-accountability work that predates the current AI wave. We'd rather you ask these questions of anyone you're evaluating — including us.

// FROM THE LAB

Real Builds, Real Numbers

Two of our own builds, with the real architecture decisions behind them.

01 · CUSTOMER SUPPORT AGENT

73% Ticket Deflection in 6 Weeks

E-commerce SME, 200–400 daily inquiries. RAG + LangGraph + HITL escalation, with the architecture decisions behind each one. Full case study →

02 · MULTI-AGENT SYSTEM

6 Months Rebuilding IRIS from Scratch

The supervisor-specialist rebuild behind IRIS v2 — why it replaced the original single-agent design, and what changed. Full case study →

FAQ

// 10 QUESTIONS

These are the questions we actually get asked — by plant managers who've been burned by an RPA sales pitch before, and by procurement teams who need a straight answer on compliance before they'll even take a second meeting. No hedging, no "it depends" without an actual answer attached.

What is AI automation?
AI automation uses software agents to handle tasks that used to need a person watching, deciding, and acting — reading a document, flagging a defect, routing an approval. See our breakdown of what separates an agent from a chatbot.
AI automation vs. traditional RPA — what's the real difference?
RPA replays fixed click-by-click scripts and breaks the moment a screen or data field changes. AI automation uses agents that read context and keep working when the input isn't identical to yesterday's. See our tooling comparison.
Do we need to replace our existing systems?
No. Our approach integrates with your existing cameras, ERP, or approval workflows rather than requiring a rebuild — see how we structure that integration layer.
How is this different from hiring more staff or buying traditional software?
Traditional software follows fixed rules and breaks on exceptions; more headcount doesn't scale linearly. AI automation handles variation and scales without proportional cost, often via several specialised agents rather than one do-everything model.
What does computer vision QC actually detect on a Penang production line?
Defect classes trained on your actual production line — not a generic off-the-shelf model — covering visual defects, dimensional checks, and barcode/QR verification. Full breakdown →
Is this PDPA-compliant?
Yes — our architecture is built around Malaysian data-residency and PDPA requirements from the start, not retrofitted after deployment. See our PDPA breakdown, or our healthcare-specific PDPA work for sensitive personal data.
How does IRIS keep a human in the loop on approvals?
IRIS surfaces information, flags risk, and enforces SLA timers — it doesn't make the final call. Every approval, escalation, and override is logged. See our human-in-the-loop design breakdown.
What's the difference between an AI agent and a RAG-based knowledge assistant?
An agent acts — it executes steps in a workflow. A RAG system answers — it retrieves and reasons over your own documents. JomWork is built on the RAG side; see our RAG pipeline breakdown.
How long does deployment take?
Depends on scope — a single-workflow pilot moves faster than a multi-department rollout. Our customer support agent project log walks through a real build end-to-end.
How much does AI automation cost in Penang?
Cost scales with the number of workflows, integrations, and deployment model — cloud, hybrid, or on-premise. We scope every engagement individually rather than publish a fixed rate card. See our AI reporting agent breakdown for how we think about measuring ROI once live.

Tell us where the friction is on your production line — we'll tell you exactly how we'd rebuild it.