MGTLTech LabsStart a Technical Brief

Building at the edge.

We build systems
before they become obvious.

AI · Edge · Automation · Data

We're not a software agency. Muai Gemilang Tech Labs is a Malaysian R&D lab — we build under the constraints Malaysian organizations actually operate in, then ship what survives contact with them.

Start a Technical BriefExplore the Lab
  • 2019Incorporated in Malaysia
  • 8Products in the lab
  • 100%Client IP ownershipPer contract

Why Muai Gemilang

We exist to push the frontier of what’s possible with technology in Malaysia — not just implement what already exists.

The lab model is different from an integrator’s. We don’t start with yesterday’s best practice. Everything we bring you has already been built, broken, and fixed — in Malaysian conditions, under real constraints.

Edge First

We don’t follow trends. We test what comes after.

R&D Driven

Everything we ship started as an experiment.

Pioneering Applications

We apply the latest technology in ways others haven’t tried yet.

Malaysian Tech Ambition

Building world-class systems from Malaysia.

Real Implementation

Not just research. We ship and maintain.

Commercial / Market Products

6 products, one named constraint each

Margin Intelligence

MarginOS

Margin intelligence for hardware distributors.

Problem
Fragmented margin visibility across principals and currencies.
Domain
  • AI
  • ERP
  • B2B
  • Analytics
In Development
Inspect build sheet

Clinic Operations

Clinic Ops

Done-with-you clinic operations, not self-serve SaaS.

Problem
Self-serve clinic SaaS stops at the login screen, not at payroll or inventory.
Domain
  • Healthcare
  • PDPA
  • Payroll
  • Edge

90msp99 latency (from 480ms)

EXP-002 · shipped, verified

In Development
Inspect build sheet

Content Platform

Bacalah

PWA web-novel platform, built for Bahasa Malaysia.

Problem
Bahasa Malaysia readers have no PWA-native, author-first web-novel platform.
Domain
  • PWA
  • Bahasa Malaysia
  • Publishing
  • Payments
In Development
Inspect build sheetVisit site

HR Compliance

RoleScope

The "first mile" hiring engine for Malaysian founders.

Problem
First-time founders hire without knowing which Employment Act clauses apply.
Domain
  • AI
  • Legal
  • Bilingual
  • Compliance
In Development
Inspect build sheet

Data Privacy

FormShield

PDPA form-compliance, not a lawyer.

Problem
Web forms collect personal data with no PDPA consent trail.
Domain
  • PDPA
  • Compliance
  • Forms
  • Audit
In Development
Inspect build sheet

Quant Infrastructure

Base Backtest

Constraint-driven backtesting for Base L2 DEX trading.

Problem
Backtests that assume idealized fills are a story, not a system.
Domain
  • DeFi
  • Base L2
  • Backtesting
  • Risk
In Development
Inspect build sheet

Lab Infrastructure

2 systems the lab runs on, not sold

Not products for a customer — the tooling this lab's own work depends on.

Agent Memory

ai_brain

Vault-grounded semantic memory for AI agents.

Problem
AI agents need fast, grounded recall over the lab’s own knowledge base.
Domain
  • pgvector
  • Embeddings
  • Retrieval
  • Self-hosted

<1swarm recall latency

Production
Inspect build sheet

Document Format

LTF

A bitemporal claim-store file format, benchmarked in public.

Problem
Claim-based document models are unproven against plain JSON on disk.
Domain
  • Bitemporal
  • Provenance
  • Benchmarking
  • Encoding

11.2%disk reduction (vs minified JSON)

EXP-008 · re-verified benchmark

Prototype
Inspect build sheet

8 builds total across both tracks — every status above is self-reported and updated as the work moves.

Build Log

We show you the build log before we show you the invoice.

Every shipped feature starts as an experiment. We publish the hypothesis, the failure, the correction and the result.

Experiment #008

Learned

A reported disk-savings "win" turned out to be measuring the wrong baseline

Hypothesis

LTF's binary format beats JSON on disk size for structured records — first benchmark showed an 11% win.

  1. First result

    Fail:11% (retracted)

    That win was measured against a pretty-printed JSON fixture, not a fair one. Against minified JSON, the format was actually larger — a wash, not a win.

  2. Correction

    Profiling where the bytes actually went — not guessing — found two real, fixable causes.

  3. Retest

    Pass:-11.2%

    Correcting them turned the retracted win into a genuine, re-verified 11.2% reduction against the fair baseline.

Lesson

An unverified win is worse than a published loss — it just fails later, in front of someone else. Every efficiency number this project publishes now states the exact baseline it was measured against.

Experiment #001

Learned

General-purpose LLM vs constrained architecture for compliance agents

Hypothesis

A generic LLM can draft MyInvois e-invoice payloads end-to-end.

  1. Result

    Fail:

    It hallucinated a permit number on a real test invoice.

Lesson

We needed a constrained architecture — not a bigger model.

Experiment #002

Shipped

Monolithic vs edge-native architecture for clinic ops latency

Hypothesis

A single Node service handles Malaysian clinic peak-concurrency fine.

  1. Result

    Pass:480ms → 90ms p99

    Edge-native moved the slowest call from 480ms to 90ms p99 without rewriting the core.

Lesson

The risk lives in the lab, not in your production.

Read the full build log — 10 logged experiments →

Where to start

What are you here for?

I have a difficult problem.

Tell us the constraint. We’ll tell you whether we’ve already solved something similar.

Start a Technical Brief

I want to see what MGTL is building.

6 products in active development, 2 systems the lab runs on internally — every status self-reported.

Explore the Lab

I want to understand the R&D.

Every shipped feature traces to a logged experiment — hypothesis, result, lesson — public before you have to ask.

Read the build log