The Understanding Enginefor Real-World AI

AI Speech & Video Solution for Southeast Asia

AI video localization & AI meeting notes

AI video localization & AI meeting notes

AI video localization & AI meeting notes

in Southeast Asia

in Southeast Asia

in Southeast Asia

AI VIDEO LOCALIZATION

One upload.
Every market.

One upload.
Every market.

One upload.
Every market.

Translation, Casting, Dubbing, Lip-Sync - One Pass

Learn more
0:00 / 0:00
0:00 / 0:00

AI MEETING NOTES

Talk.
It writes.

Talk.
It writes.

Talk.
It writes.

Upload audio, or record live, and get structured
minutes — decisions, action items and owners —
with speakers separated and word-level timestamps.

Upload audio, or record live, and get structured minutes — decisions, action items and owners — with speakers separated and word-level timestamps.

Learn more

YOUR LANGUAGE · YOUR INDUSTRY · YOUR RULES

AI, trained on your world.

Arclek Reactor makes it real — a pipeline of three engines that turns scarce data into training corpora,
deep domain logic into a reasoning graph, and every answer into one that’s been verified.

THE CHALLENGETHE OUTCOME

Arclek Reactor

THE UNDERSTANDING ENGINE

Misreads real speech.

Accents, code-switch, dialects.

Speaks your language.

Trained on data
grown from yours.

Answers generically.

Knows the internet,
not your industry.

Knows your business.

Your terms and rules,
wired for reasoning.

Sounds confident.

Nothing checks it.

Proven before it ships.

Every answer graded
to your standards.

About us

About us

Arclek is an AI company building speech and knowledge AI for Southeast Asia. Its models rest on the team's own synthetic data engine and context graph — methods the founders authored and published — which is how it reaches state-of-the-art accuracy in Thai speech recognition and builds AI that reasons inside a specific business rather than the internet at large. Arclek ships two products, ArcDub for AI video localization and ArcNotes for AI meeting notes, both deployable inside the client's own infrastructure and extendable through productized customization.

WHY

Southeast Asian languages are a rounding error in the web corpora that open pretraining sets are built from: in Common Crawl's latest crawl, English covers 42.6% of pages and Thai 0.35%. A general open model reaches only 82.99% word accuracy on Thai, against 92.79% for Arclek's Thai speech recognition — meaning is lost before any reasoning starts. Arclek trains speech models for these languages specifically, rather than inheriting whatever a global model happened to absorb.

Common Crawl CC-MAIN-2026-43 · share of pages by primary language

An industry's working knowledge — product clauses, compliance scripts, decision paths, edge cases — lives in internal documents and in veterans' heads, not on the public internet. No amount of prompting retrieves what was never in the training data. Arclek structures a client's own material into a context graph, so the model reasons inside their business instead of guessing at it.

HOW

A synthetic data engine turns a small amount of real material into training-scale data. Arclek's engine grows a client's seed corpus 10–100×, which is how models get built for languages and industries where real data is scarce, sensitive or long-tail. The client's raw records stay sealed throughout; only the patterns are learned.

A context graph structures an organization's terms, rules, decision paths and edge cases so a model can reason over them rather than merely retrieve matching text. The method is published research, and Arclek runs it in production knowledge and agent systems. It is what makes an AI answer like someone who has worked in the business, not like a search box.

Arclek's Thai speech recognition reaches 92.79% word accuracy, ahead of every global vendor measured. The same engine, applied to Malay, reaches 87.6% on an in-domain test set and ranks first on FLEURS at 7.13% WER.

Arclek ASR-TH

92.79

ElevenLabs Scribe

91.68

Google Chirp v3

91.44

Whisper-large-v3

82.99

Qwen3-asr

68.19

Thai ASR · word accuracy % · internal evaluation, measured Q3 2026 · results may vary

Arclek trains its own speech models for this region rather than relying on a global model that happens to include these languages. The work so far is concentrated on Thai and Malay — Thai tuned and measured first against every global vendor, Malay reached by the same engine. Other Southeast Asian languages run the same path: a small seed corpus grown into training data, then a model trained on it. Which language comes next is decided with the client who needs it.

PRODUCTS & WORKING TOGETHER

In beta from Q4 2026. ArcDub is Arclek's AI video localization product: one upload becomes a finished localized episode — speech recognized, subtitles translated and timed, voices dubbed and lip-aware. Character voices, glossary terms and subtitle styles carry across a whole series, and editing a single line re-renders only that segment. Speech recognition runs on Arclek's own models, trained for Southeast Asian languages.

In beta from Q4 2026. ArcNotes is Arclek's AI meeting notes product: upload audio or video, or record live, and get structured minutes — decisions, action items and owners — with speakers separated and word-level timestamps. The same transcription is available to enterprise customers through the Arclek API.

Yes — on top of ArcDub and ArcNotes. Productized customization covers private deployment inside your own infrastructure, tuning for a specific language, and strengthening specific features, with the work scoped to what the product roadmap can carry. Tell us what you need here and we will scope it with you.

COMPANY

Arclek is the international brand of DataArc Technology Limited, a Hong Kong company founded in 2025 and incubated by IDEA Research Institute, founded by Dr. Harry Shum, former Executive Vice President of Microsoft. Dr. Xuhui Jiang, Founder and CEO, holds a PhD from the Institute of Computing Technology at the Chinese Academy of Sciences and was previously a research scientist and head of knowledge-graph R&D at IDEA. Dr. Chengjin Xu, CTO, holds a PhD from the University of Bonn, was named a Huawei "Genius Youth", and previously led IDEA's financial large-model programme. The team is 80% PhD or Masters, with 70+ papers, 10,000+ citations, 3,000+ GitHub stars and 10+ patents. The methods behind the pipeline — Think-on-Graph, Synthesis-on-Graph and LLM-as-a-Judge — are published, and the toolkits are open-sourced on GitHub.

BUILT AROUND YOUR BUSINESS

Talk to our AI
solutions experts.

Talk to our AI
solutions experts.

Talk to our AI
solutions experts.

Interested in

Speech and knowledge AI for Southeast Asian enterprises.

© 2026 DataArc Technology Limited.