META COMMUNICATION CENTERThe quality of a conversation, diagnosed by AI.
Meta Communication Center builds AI that analyzes and corrects Korean speech. From cross-generational understanding to speech training to safety control over what AI and public figures say — we raise the quality of every moment a message is delivered.
We work on how people speak
In an age of text without a face or a tone of voice, misunderstanding starts with a small turn of phrase. We took seventeen years of domain assets — broadcast newsrooms, and speech coaching for senior public officials and executives — and built them into a speech-judgment AI specialized for Korean.
One judgment engine serves many fields at once; only the rubric changes. An advisory panel of PhDs in Korean linguistics reviews the quantitative reliability of those judgments on a regular schedule.
One founder set the standard. The other built the engine.Seventeen years of domain judgment met frontier-lab LLM safety engineering inside one company.
An eye that reads speech, and a hand that turns that reading into a system — the IP only works when both are present.
CEOHyeonggi Kim
Speech-context analysis · rubric design- 17 years in Korean broadcast news — MBC newsroom, YTN anchor desk, TV Chosun, SBS
- 600+ speech-risk coaching engagements for senior officials and executives
- Korean patent 10-2774556 (speech-diagnosis-based learning) · two registered copyrights
- Distinguished instructor, National Human Resources Development Institute · 2025 Education Innovation Leader Award
Holds the core rubric-design IP CTOJaewoo Kim
LLM control · architecture for highly regulated AI- Former LLM engineer at a frontier AI lab — safety and hallucination control
- Infrastructure design for national security networks and Korea’s three major telecoms
- Judge-LLM architecture and guardrail pipeline design
Frontier-lab LLM safety engineering Combined IPThree products, one engine
Generational gapInterAI
Interpreter + AIThe same Korean sentence lands differently across generations. InterAI reads tone, context and generational idiom to interpret what was actually meant — in both directions — and scores how your words register with the other side.
interai.kr ↗Speech trainingKORHUB
Personalized Korean speech trainingA mobile speech-training platform combining AI speech analysis (stability, accuracy, delivery, pace) with one-on-one consulting from former broadcast anchors. Built for corporate staff, instructors and the general public.
korhub.co.kr ↗AI safety controlDRG
DeepRegGuard · speech safety control layerAn external safety layer that inspects what AI systems and people are about to say, immediately before it reaches a customer or the public. Violations, hallucinations and reputational risk are blocked or corrected in real time.
Partners only · contact us INTERAI LENSGlasses that decide what your eyes need
InterAI Lens is captioning and speech-coaching software for heads-up display glasses that have no camera and no speaker. Existing captioning glasses — Xander Glasses, AirCaps and others — transcribe what they hear. InterAI Lens runs a judgment engine that decides which words earn the two lines of available screen space.
Access modeSurrounding speech and non-speech sounds become text in the wearer's line of sight, so deaf and hard-of-hearing users can follow a conversation without looking down at a phone. Background conversation is suppressed; a question aimed at the wearer is promoted with a marker.
Coach modeThe same engine evaluates the wearer's own speech in real time and shows a corrected phrasing. Many deaf users cannot hear their own articulation — a barrier no captioning product addresses.
Notification budgetPrecision over recall. For a wearer, a flood of captions is worse than a missing one. What the engine withholds is as much the product as what it displays.
Privacy by hardwareNo camera, no speaker. Personally identifiable information is re-masked server-side, and raw audio is discarded after transcription.
The engine stays. Only the standard changes.
Not a promise to build one — a judgment engine already running in public service. Speech moves through a four-stage pipeline.
01High-speed first filterRule-based immediate blocking. Most input is handled without a language-model call, which keeps cost and latency down.
02Real-time fact checkCross-referenced against verification data to surface hallucinations and errors.
03Contextual decisionMulti-agent consensus judges context and intent, and reports the reasoning behind the call.
04Audit log retentionThe grounds for each judgment are recorded and retained in tamper-evident form.
Verified technology and credentials
Registered patentKorean patent 10-2774556 — learning technology based on AI speech diagnosis
Seoul Fintech Lab residentA vetted technology company within Korea's fintech ecosystem
Financial regulatory sandboxSelected for a Financial Services Commission consignment test — evidence of technical novelty and necessity
Academic advisory panelRegular quantitative review by PhDs in Korean language and linguistics
17 years of domain assetsBroadcast newsrooms plus a speech-coaching archive for senior officials and executives
Eight live domains (and growing)One engine carrying transplanted expert judgment, coaching high-difficulty speech around the clock — public and private apologies, insurance advisory, citizen complaint handling and more