VietCheckMed AI

01 — The Crisis

Vietnam's Medical
Misinformation Epidemic

Unregulated medical ads flood social media daily, hiding illegal surgical acts behind compliant text — causing irreversible health damage and death.

📱

Digital Explosion

Unregulated medical advertisements flood social media platforms daily at scale, far outpacing any manual review capacity.

🎭

Cross-Modal Evasion

Safe, compliant text conceals illegal surgical procedures and prohibited acts within visuals — bypassing all text-based AI moderators.

⚗️

Complex Deception

Pseudo-scientific claims and fabricated medical authority manipulate consumer trust with sophisticated, constantly evolving tactics.

⚡

Zero-Day Regulatory Lag

Vietnamese jurisprudence changes rapidly, but traditional AI requires months of retraining — leaving legal gaps that fraudsters exploit immediately.

Formal Healthcare Gap

Rigid processes and strict compliance standards are slow to react against fast-evolving digital misinformation that operates in the dark with highly unstructured data.

Black-Box Accountability Crisis

Inspectors cannot penalize clinics based on AI confidence scores alone — every compliance decision demands legally grounded, auditable explanations.

02 — Our Solution

Not an App. A Full
Intelligence Ecosystem.

VietCheckMed is the only platform combining multimodal AI, legal-grade evidence, and proactive surveillance to counter medical fraud before it causes harm.

🛡️

Transparent AI

Not a Black Box

Every verdict is explainable and traceable. SBV-certified evidence standards mean every conclusion is legally defensible in Vietnamese courts.

⚡

Proactive Intelligence

Not Reactive. Proactive.

With Watchtower Agent & Graph Analytics, VietCheckMed detects, traces, and dismantles fraud clusters at inception — not after harm is done.

🌐

Full Stack Platform

Ecosystem Approach

Microservices + Kafka architecture supports sub-3s public responses, enterprise-grade stability, and deep regulatory analytics simultaneously.

vs. Global WHO EARS

Passive monitoring, text-only

vs. Health RAG Frameworks

Text-centric, no compliance check

vs. Domestic Tools

Manual, unscalable, zero AI

VietCheckMed

Proactive · Multimodal · Legal-grade

03 — Core Workflows

One Platform. Three Shields.

VietCheckMed serves three distinct segments — each with specialized workflows tailored to their unique compliance needs.

Community Reporting

User reports a suspicious phone number or image from an unverified medical ad.

Builds crowdsourced intelligence dataset; adds contact to active "Under Monitoring" list.

Instant Verification

User pastes a URL, phone number, or app name into the Web/Mobile interface.

Instant WARNING in under 30 seconds — blocks fraud at the zero moment of decision.

Safe Discover

User searches VietCheckMed's Verified Medical Directory for trustworthy providers.

Returns list of verified legitimate clinics and practitioners for safe navigation.

⚡

Speed requirement: Sub-10s response with lightweight inference pipelines — designed for zero-friction consumer access.

Active Monitoring

Watchtower Agent automatically scans new domains and flagged fanpages continuously.

Intelligence dashboard with early warnings and fraud cluster analysis for regulators.

Investigative Support

Inspector inputs "seed data" — phone number, domain, or bank account — to begin investigation.

Knowledge graph for multi-source linkage and deep tracing of fraud networks.

⚖️

Regulatory requirement: Multi-hop reasoning, deep retrieval, and formally cited statutory violations for legally defensible enforcement.

Ad Screening API

Partner platforms (ad networks, social media) call the API for every new medical advertisement.

Returns risk score with automated approval/denial decision and detailed evidence log.

Brand Protection

Organizations register official channels; system continuously scans for impersonations.

Verification badge issued + real-time impersonation alerts to protect brand integrity.

News Verification

Journalists verify apps, phones, and websites before publishing investigative reports.

Detailed source verification report for rigorous, evidence-backed fact-checking.

🔌

API requirement: 24/7 uptime with instant adaptation to new regulatory decrees — enterprise-grade stability built in.

04 — Technology

Four Specialized AI Agents
Working in Concert

MM-RegCheck operates through four sequential AI modules — each purpose-built to handle a different dimension of multimodal compliance reasoning.

👁️

VLM · Agent 1

Visual-Clinical Extractor

Vision Language Model that detects clinical acts embedded in images and video, bypassing semantic camouflage that fools text-only moderators.

📚

Agentic RAG · Agent 2

Statutory Retriever

Matches extracted medical acts against the permitted scope of each clinic from Vietnam's Regulatory Corpus using lexical and semantic search.

⚖️

TF-BEAR · Agent 3

Causal Adjudicator

Calculates causal deltas between text and visual evidence — overriding misleading compliant text when illegal procedures are detected in images.

📋

XAI · Agent 4

Explainable Generator

Produces formal Vietnamese compliance reports citing exact statutory violations — delivering legally defensible verdicts for enforcement action.

MM-RegCheck Pipeline

🖼️

Extraction

Text + Image Input

↓

🔍

Retrieval

Regulatory Corpus

↓

⚡

TF-BEAR Debias

Prior-Posterior Reasoning

↓

📄

Explanation

Statutory Verdict

Architecture resolves B2C/B2G/B2B conflicts through three core principles:

Isolation

🚦

Functional Microservices

Fast lane (B2C) is isolated from heavy analytical lane (B2G), ensuring no latency bleed between segments.

Transport

⚡

Asynchronous Kafka

Continuous streaming from crawler, social monitoring, and user reports — events processed in real time at scale.

Evidence

🏛️

Legal-Grade Truth Layer

SBV whitelist, standardized blacklist, and legal-grade evidence archives form the authoritative ground truth.

05 — Scientific Validation

Evidence-Based Performance

MM-RegCheck was rigorously evaluated against state-of-the-art baselines across standard, ungrounded, and adversarial conditions.

SOTA performance across all evaluation conditions

Beauty & Healthcare domains

~6K

Annotated samples in MedReg-MM dataset

Medical / Aesthetics compliance

3×

FIDES-Score dimensions: Fidelity, Justification, Clarity

Explanation quality metrics

<10s

B2C inference latency with lightweight pipeline

Sub-3s for public responses

Benchmark Dataset Comparison

MedReg-MM is the first dataset purpose-built for Vietnamese medical compliance verification.

Dataset	Domain	Verification Target	Samples	Adversarial Subsets
Fakeddit	Reddit / Social Gossip	Fake News Classification	~1M	—
Factify	General News	Veracity & Entailment	~50K	—
MOCHEG	Politics / News	Veracity	~15K	—
MMM-Fact	Multiple domains	Veracity	~125K	—
MedReg-MM Ours	Medical / Aesthetics	Compliance (Authorization)	~6K	✓ Hard Evasion

06 — Deployment Roadmap

From Pilot to National Scale

A three-phase deployment strategy — from HCMC pilot to 24/7 nationwide automated surveillance.

Phase 1

Months 1 – 6

MVP Pilot

Launch Web App & Browser Extension with sub-10s inference. Partner with HCMC Health Department for regulatory pilot validation.

Phase 2

Months 6 – 12

Scale & Beta

Expand to nationwide regulatory corpus. Launch public beta with awareness campaign. Onboard B2B API partners (Zalo, MoMo, Ad Networks).

Phase 3

Year 1+

Auto System

24/7 autonomous surveillance with full SaaS integration. Automated detection, reporting, and enforcement pipeline for national regulators.

Demo

Our Commitment

Trust Built on
Evidence-Based Data

VietCheckMed is not a product — it is a commitment to a safer Vietnamese healthcare ecosystem. Every result is explainable. Every verdict is traceable. Every decision is legally defensible.

🏛️