BACEN 4.658Cybersecurity Resolution
Mandatory cybersecurity policy for financial institutions. AI models in critical pipelines require environment segregation, risk classification and documented BCP.
Applied AI for banks, financial services.
Credit, fraud, customer support and compliance — for banks, fintechs, lenders, asset managers and brokers.
Overview
What we deliver in this vertical.
Banks · Fintechs · Investments
Financial Services
Credit, fraud, customer support and compliance — for banks, fintechs, lenders, asset managers and brokers.
−38%credit loss
−72%analysis time
+24%credit conversion
Applications
- 01Alternative credit with open finance
- 02KYC and fraud with generative AI
- 03Regulated support (LGPD, central bank)
- 04Document analysis of contracts
- 05Copilots for trading desks
Positioning
AI in finance isn't optimization — it's competitive baseline.
Banks still treating AI as innovation projects are losing ground to those treating it as infrastructure. Open finance, instant payments, BNPL and real-time fraud don't tolerate analog pipelines. The gap between a versioned, monitored, auditable scoring model and a spreadsheet with fixed rules is the entire spread of a portfolio.
By the numbers
What matters when the credit committee meets.
−0%PJ credit loss in 12 months, with stable approval
−0%average KYC time with AI document analysis
+0%credit conversion lift, default rate flat
<0msmedian real-time scoring latency
Under the hood
A stack that survives audit.
Each component is vendor-neutral and chosen to satisfy Brazilian central bank regs, LGPD and — when applicable — PCI-DSS.
Models
- Self-hosted Llama / MistralDefault for customer data in regulated environments.
- Claude (BAA)When the commercial contract covers the case.
- Fine-tuned proprietary modelsScoring, fraud, transaction classification.
Data and features
- Open Finance ingestionReal-time pipeline with LGPD-compliant retention.
- Feature store (Feast / Tecton)Versioning + lineage to raw data.
- PIX & ACH streamKafka + event processing.
Governance
- Versioned model-cardsIn central-bank/ANPD format.
- Signed logs (HSM)Each block/approval reproducible.
- Drift monitoringAlerts before the KPI degrades.
Security
- Isolated VPC + granular IAMNo data leaves the perimeter.
- PII tokenizationPCI-DSS compliant.
- DPA + BAA signedDefault for any commercial model.
Compliance
Frameworks your model must respect — before the first POC.
The regulatory journey starts before architecture. Each financial use case crosses at least three parallel frameworks. We treat them as production gates, not end-of-project checklists.
LGPDBrazilian General Data Protection Law
Purpose-bound legal basis, minimum retention, right to explanation in automated decisions. We run DPIA before training and version RoPA per model.
PCI-DSSPayment Card Industry
When the pipeline touches card data (even encrypted), we require tokenization, network segmentation and SAQ-D audit. Default for any transactional fraud use case.
BCB · AIBrazilian Central Bank AI Roadmap
BCB guidelines for AI in financial markets: governance, transparency, risk management. Model-cards and decision documentation prepared for regulatory review.
Open FinanceBrazilian Open Finance Convention
Granular consent, limited retention and mandatory revocation. Models consuming open finance data need a complete trail from consent to decision.
In production
Results in financial services.
Alternative credit · mid-size bank
Scoring model leveraging open finance reduced PJ credit loss 38% in 12 months without lowering approval.
AI KYC · fintech
Automatic document analysis cut average KYC time from 4h to under 60min. Compliance kept at 100%.
Credit conversion · retail
Per-channel and per-moment credit offer personalization. Conversion lift without proportional default increase.
04 · LA AI Method
From hypothesis to operation.
Five steps that prevent eternal pilots. Each stage has deliverables, metrics and decision gates.
M1
Diagnostic
Opportunity map, estimated value, data readiness.
2–3 wksM2
Strategy
Prioritized roadmap, reference architecture, governance.
2 wksM3
Proof of Value
POC with production criteria and business metric defined.
4–6 wksM4
Deployment
Model, integrations, security, observability and UX.
6–12 wksM5
Continuous Operations
Evolution, fine-tuning, eval suite, cost per use, new features.
RecurringFrequently asked
About AI in financial services.
How do you handle Brazilian central bank and data regulators?
From day one we align governance with central bank's AI roadmap and Brazilian data law. Model-cards, explainability documentation and decision logs are part of the deliverable.
Is self-hosted mandatory for sensitive data?
When the regulator requires, yes. Otherwise we offer commercial models with BAA + DPA, depending on the institution's risk appetite.
Model audit for the central bank?
Each model has training docs, dataset, metrics and historical drift. Package is delivered in the format expected by internal or external audit.
Average time to deploy?
For a new use case: 4–8 weeks for production-criteria POC, plus 6–12 weeks for full deployment with observability and governance.