ViwisAILab

Inspiration

70% of Vietnamese MSMEs are rejected for bank loans. They have revenue, customers, and real operations — but no credit history. Banks spend 20−50permanualapplicationreview,andghostcompanies(fakebusinesses)drain 20−50permanualapplicationreview,andghostcompanies(fakebusinesses)drain2B+ annually through fraud. Vietnam's Jan 2026 e-invoice mandate just created a new data goldmine that nobody is using for credit decisions yet.

What it does

CredoAI is a credit intelligence API that turns Vietnam's newly mandated e-invoice data, bank transactions, and digital footprint into instant credit profiles. Two core products:

Ghost Detection — 5-factor parallel verification (VietQR, Google Maps, social presence, registration, reviews) catches fake companies in 5 seconds vs 4 hours manual review. Adaptive Credit Scoring — Weights shift based on data availability. A pho shop with KiotViet POS data gets scored differently from a construction company with only bank transactions. No competitor adapts to incomplete data — they all require fixed inputs that Vietnamese micro-SMEs don't have.

How we built it

Next.js 14 + TypeScript (monolith, Vercel-deployed) Supabase PostgreSQL with Row Level Security Gemini API (LLM reasoning + Vision for OCR) Casso API (bank transaction webhooks) VietQR (government bank verification) SerpAPI (digital footprint scanning) 47 microservices: ghost detection, credit scoring engine, agentic underwriter with Cedar policy guardrails, episodic memory, circular trading detection

Challenges we ran into

Accomplishments that we're proud of

Adaptive scoring engine (weights shift by data depth — no competitor does this) Deterministic fast-path (60% of cases skip LLM entirely0.02/scorevs0.02/scorevs0.50 at Experian Counterfactual hints ("upload 3 more invoices to gain ~4 points") Backer Boost (formalizes Vietnamese "tín chấp" community guarantee into scoring model) Ghost detection API live, 5-second response time Architecture: 12 active services, 33 archived for future phases Unit economics: $0.04-0.09 per API call, 85-90% gross margin Target: 70 MFIs + 35 banks + 150 fintechs in Vietnam = $50-100M SAM

What we learned

What's next for ViwisAILab

Looking for enterprise retail partners (KiotViet, SAPO, POS providers) to integrate ghost detection into merchant onboarding flows. Every free check builds our data moat — 100K checks = largest ghost company database in Vietnam.

Built With

• agentic-underwriter-with-cedar-policy-guardrails
• circular
• credit-scoring-engine
• episodic-memory
• next.js-14-+-typescript-(monolith
• trading
• vercel-deployed)-supabase-postgresql-with-row-level-security-gemini-api-(llm-reasoning-+-vision-for-ocr)-casso-api-(bank-transaction-webhooks)-vietqr-(government-bank-verification)-serpapi-(digital-footprint-scanning)-47-microservices:-ghost-detection