S01 / Seoul / Early-stage
Team AI Lab is a global AI startup for regulated clinics and enterprises: dental, medical, Korean medicine, finance, and company systems. The architecture is Frontier AI & Multi-Model Reasoning (Claude / Gemini / Grok / Local SLM) + CipherFlow Privacy Airlock. Names and resident registration numbers become pseudonym tokens on the edge before any model call.
Product-Led B2B SaaS & Enterprise AI Security Software Startup (Independent Software Vendor — Not an Agency or Consultancy). 100% of revenue comes from recurring SaaS subscriptions and enterprise software licensing. Clinic SaaS Plan $199 / month per clinic. Growth Gateway Plan $990 / month per cluster. Enterprise Sovereign Grid from $24,000 / year. Business Registration No. 544-36-01677. Founded January 30, 2026 (2026-01-30). Company & Pricing (/about)
S02 / Multi-model reasoning
CipherFlow removes identifiers on the edge. The same de-identified prompt can go to Claude, Gemini, Grok, or a local SLM. The viewer below is the Claude API path used in the clinical products: dual-agent audit on the first-party Anthropic console.
Names and resident registration numbers are replaced with pseudonym tokens on the edge before any model call leaves the premises.
Agent A looks for under-coding and omissions. Agent B answers as a HIRA notice and regulatory reviewer.
Only a specialist’s direct keypad signature is accepted. Agent hearsay is rejected.
On the clinic or company edge, the CipherFlow Defense Engine substitutes detected names and resident registration numbers in full with pseudonym tokens before the next step. The engine is Korean GLiNER NER, an RRN checksum validator, and a clinical allowlist.
A de-identified prompt with zero bytes of raw PII is what the first-party Claude API receives. Agent A (under-coding and omission discovery) and Agent B (HIRA notice and regulatory reviewer) run long-form reasoning and tool use against each other.
Only a specialist’s direct keypad signature (DIRECTOR_KEYPAD) — a clinic director or reviewer — is accepted. AGENT_HEARSAY is blocked. The remaining outputs are a PII-free audit record and a personalization calibration vector.
S03 / Four deployed surfaces
Four deployed surfaces a reviewer can open now. This page does not use customer logos or clinical accuracy scores. The test counts below are the startup-review inventory, not a clinical performance score.
| Product | Live surface | What is deployed | Automated tests |
|---|---|---|---|
| CipherFlow | https://www.teamailab.com | Core Privacy Airlock gateway and Defense Engine. Go Gateway, Python FastAPI, Korean GLiNER, RRN checksum guard. | 59 automated tests passing |
| DentGuard | https://dentguard.teamailab.com | Dental HIRA pre-audit local blackbox. 2-track EDI SAM and chart ingestor, 15 public HIRA dental rules, phase-1 retrospective backtest runner. | 40 automated tests passing |
| MediGuard | https://mediguard.teamailab.com | Medical clinic HIRA pre-audit appliance. C-arm, chronic care, endoscopy, and physical therapy — 5 scenarios. | 13 automated tests passing |
| HaniGuard | https://haniguard.teamailab.com | Korean medicine clinic HIRA pre-audit appliance. Chuna, auto insurance, and herbal medicine — 5 scenarios. | 13 automated tests passing |
The gateway and defense engine are deployed. This tab shows that the link is a live surface and which controls are in scope. There is no throughput slogan.
Dental pre-audit local blackbox. The public notice catalog and the retrospective backtest runner are in scope.
Medical clinic pre-audit appliance. Five scenarios are the current scope.
Korean medicine pre-audit appliance. Five scenarios are the current scope.
S04 / Architecture evidence
This table is the architecture artifact a reviewer can check against the repositories. Do not read it as a clinical F1 score or a customer-adoption claim.
| Artifact | What it is | What it is not |
|---|---|---|
| 2-track non-intrusive collection | EDI SAM and chart ingest do not replace the clinic system of record. | Not a claim that every EHR integration is finished. |
| deterministic pseudonym join | A stable join key after local tokenization. | Not a reversible patient identifier sent to the model. |
| 15 public HIRA dental rules | The public notice catalog used by DentGuard. | Not a private or complete claims database. |
| Clean-room compatibility | The model sees a de-identified prompt. Raw PII stays on premises. | Not a certification badge or a third-party warranty. |
This page is not an Anthropic endorsement, partnership, or certification.
S05 / Founder
Leads enterprise mission-critical architecture, QA gates for highly regulated systems, and AI agent engineering.
Seoul, South Korea
S06 / Startup verification
Operating identity for partners, customers, and program reviewers. This is the company record, not a page tied to one application program.
| Field | Value |
|---|---|
| Company Legal/Operating Name | Team AI Lab |
| Domain | https://teamailab.com |
| Work Email Domain | @teamailab.com |
| Founded | January 30, 2026 (2026-01-30) |
| Headquarters | Seoul, South Korea |
| Funding Status | Bootstrapped / Pre-Seed |
| Model architecture | Frontier AI & Multi-Model Reasoning (Claude / Gemini / Grok / Local SLM) + CipherFlow Privacy Airlock. The Claude path is adversarial dual-agent audit, policy evaluation, and agentic engineering. |
The corporate site is static. Language choice is stored only in the localStorage key teamailab-lang on this browser. There is no analytics pixel, account, or payment form. Product appliances are separate deployments.
Each model call follows that provider’s terms. The Claude API follows Anthropic’s terms. Team AI Lab does not describe itself as an agent or certified partner of any model company. Product descriptions are architecture controls, not a promise of clinical outcome or claim approval.
Contact: david@teamailab.com