naviai Platform · Intelligent Risk

Real-time credit decisions,
governed end to end.

An intelligent risk-management platform for consumer and inclusive lending. Built on AI, big data, cloud computing and microservices around the core credit-risk workflow, it provides risk technology services to the ecosystem that matches C-end borrowers with B-end funders.

  • Decision engine
  • Feature platform
  • Model platform
  • Compute engine
Book a risk demo See the architecture →
Latency for complex real-time variables
≤0ms
In-house decision-flow node types
0+
Decoupled layers: data, features, models, decisions
0
Lineage across features, models and flows
0%

Application & business model

A fintech ecosystem: C-end borrowers — risk platform — B-end funders.

Through one unified interface and task-scheduling layer, the platform serves both online real-time risk control and offline batch analysis — and gives funders value-added services in strategy iteration, risk monitoring and post-loan management.

C-end borrowers

  • Consumer finance
  • Inclusive finance

Intelligent Risk Management Platform

AI · Big data · Cloud · Microservices

  1. Data ingestion
  2. Feature computing
  3. Model scoring
  4. Decision output
  5. Monitoring & alerts

B-end funders

  • Banks / licensed institutions
  • Consumer finance companies
  • Trusts / microlenders
  • Online lending platforms

Core capabilities

  • End-to-end integration
  • Visual decision flows
  • Real-time + batch
  • Explainable models
  • Full-chain lineage

Product traits

  • Technically advanced
  • Feature-complete
  • Responsive
  • Stable in operation
  • Replicable at scale
01 · DECISION

Unified decision engine

Rules, scorecards, decision trees and model strategies as components, orchestrated visually.

02 · FEATURE

Feature platform

Full feature lifecycle management, version control, audit trail and variable services.

03 · MODEL

Model platform

A closed loop from training and evaluation to versioning, approval, deployment, inference, monitoring and maintenance.

04 · COMPUTE

Compute engine

Unified real-time and batch scheduling for high concurrency, low latency and many parallel tasks.

05 · END TO END

Full-chain integration

Data ingestion → feature computing → model scoring → decision output → monitoring and alerts.

Construction goals

One risk system across the whole credit lifecycle.

We set out to build a technically advanced, feature-complete and responsive risk platform that combines big data, AI and cloud computing into intelligent risk control across every stage of lending. Decision engine, feature platform, model platform and compute engine come together in one solution — sharpening risk detection, decision speed and business fit, and forming a data-driven risk ecosystem centred on customer-lifecycle management.

Decision
limit & pricing

Internal admission

  • Age not eligible
  • High-risk region
  • Internal blacklist
  • Declined by other policies
  • Poor past repayment with us

External admission

  • Court-listed defaulters
  • Third-party high-risk lists
  • Poor repayment on past loans

Customer segmentation

  • Education
  • Account age
  • Channel
  • Other

Full-stack anti-fraud

  • Face recognition
  • Relationship graph
  • Distributed graph computing
  • Multi-dimensional ML anti-fraud
  • Anti-fraud models

Credit scoring

  • Past borrowing and repayment behaviour
  • External third-party data

Credit review

  • Adverse credit information
  • Credit-bureau decline rules (status, bureau, social, overdue, over-indebtedness, cash-hungry)
  • Overall credit risk

Limit & pricing

  • Income model (bureau and third-party data)
  • Limit model (income model + overall credit rating)
  • Limit cuts for high risk
  • Risk-based pricing model

PRE-LOANRisk up front — stop it at the source

  • Blacklist screening
  • Full-stack anti-fraud
  • Customer value model
  • Automated decisions

IN-LOANTrack during the loan — manage it well

  • In-loan monitoring
  • In-loan management

POST-LOANDynamic collections — keep the customer

  • Segment analytics and retention
  • Efficient early collections
  • Collection strategies
  • Big-data skip tracing

Overall technical architecture

Four layers working as one: big data + AI + cloud.

A layered, decoupled microservice architecture. Every module connects through standard APIs, real-time and batch pipelines run side by side, and the same stack serves online risk control and offline batch analysis.

DECISIONDecision layer

Rule management · Decision-flow orchestration · Real-time decisions · Testing & validation · Strategy monitoring

MODELModel layer

Model training · Model registry · Model deployment · Model monitoring · Model explanation

FEATUREFeature layer

Feature development · Feature management · Version control · Audit trail · Variable services

DATAData layer

Multi-source ingestion · Data cleansing · Data governance · Data storage

Modules connect through standard APIs · Real-time and batch pipelines in parallel · Online real-time risk control and offline batch analysis on one stack

ROUTE

Technical route: four decoupled, cooperating layers

Data layer

  • Compute engines for real-time
  • and batch data processing
  • Flexible multi-source ingestion

Feature layer

  • Standard feature-service APIs
  • Decoupled from model and decision layers
  • Faster iteration

Model layer

  • Self-training + inference acceleration
  • Model updates need no change to decision logic

Decision layer

  • Visually configured decision flows
  • Strategy changes without code
  • Shorter iteration cycles
Four-layer
decoupled
architecture

Platform architecture

Everything a credit-risk team runs on, in one platform.

From the data warehouse up through compute, features and models to the decision engine — each layer is a product in its own right, and together they form one governed system.

Decision
engine
Strategy components
  • Rule sets
  • Decision trees
  • Decision tables
  • Decision flows
  • Scorecards
  • Champion / challenger
Strategy routing
  • Traffic split
  • Grouping
  • Parallel
  • Aggregation
Engine management
  • Version control
  • Hot deployment
  • Test & backtest
  • Component dependencies
Monitoring centre
  • Decision-flow monitoring
  • System monitoring
  • Business monitoring
  • Log alerts
Feature
platform
  • Feature management
  • Data-source management
  • Interface management
  • Function management
Model
platform
  • Model training
  • Model deployment
  • Feature management
  • Model backtesting
Compute
engine
  • Batch compute engine
  • Real-time compute engine
  • Task scheduling
Data
warehouse
  • Transactions
  • Lending
  • Credit bureau
  • Device
  • Customer
  • Third-party
  • Behaviour
  • …

System interaction

Clear boundaries: the engine talks to one data centre.

The decision engine only ever calls the variable centre. The variable centre hides every downstream difference and pulls from the model platform, the feature platform and the credit bureau — so data sources can change without touching a single strategy.

System responsibilities · click a node

    BOUNDARIES

    How the platforms divide the work

    Variable centre vs feature platform

    Variable centre
    The decision engine's data centre. It hides downstream differences and reliably supplies data to the engine, calling the model platform, the feature platform and the central-bank bureau. The engine talks only to the variable centre.
    Feature platform
    Feature configuration, derivation and computation for risk strategies, models and marketing. It is called by the model platform and the variable centre.

    Third-party data platform vs feature platform

    Third-party data platform
    The single entry point for every third-party data query in the company. It hides differences between vendor APIs and offers unified caching, querying and billing.
    Feature platform
    Connects to the third-party service to build third-party features. The two are linked: data sources register to the feature platform automatically, so nothing is configured twice.

    Credit-bureau platform vs feature platform

    Credit-bureau platform
    Triggers bureau requests, queries and caches bureau features, and switches between bureau channels.
    Feature platform
    Turns bureau results into derived features with lineage, and serves them through one unified output to models, strategies and marketing.

    Inside the platform

    Decision engine, model platform, feature platform.

    DECISION ENGINE

    Decision engine

    Manage business scenarios and rules in one place, adjust live strategies easily, sharpen risk detection and cut funding losses. Main functions: scenario management, variable centre, engine management, strategy lab, release management and call centre.

    1. Component centreRules / scorecards / decision trees
    2. Engine centreParser + decision-flow execution
    3. Test centreVisual configuration and rule testing
    4. Real-time decisionsHot deployment + online feature retrieval
    5. Query & analysisLog tracing + result backtracking
    1. Scenario intake
    2. Variable prep
    3. Decision config
    4. Decision validation
    5. Decision release
    6. Online monitoring
    Scenario management
    • Scenario definition
    • Scenario routing
    • Scenario inputs
    • Scenario outputs
    • Missing-value monitoring
    Variable centre
    Features
    • Internal
    • Model
    • Third-party
    • Central bank
    Variables
    • Derived
    • Basic
    • Constants
    Engine management
    • Decision flows
    • Flow copy
    • All rules
    • All variables
    • All I/O params
    • Version diff
    • Components
    • Rule sets
    • Component copy
    • Component check
    • Hit management
    • Front-end tuning
    Strategy lab
    • Samples
    • Offline test
    • Re-run
    • Single-case test
    • Variable calls
    • Online test
    • Multi-version
    • Component test
    • Result diff
    • Execution trace
    • Test reports
    • Data standards
    Release management
    • Canary release
    • Approvals
    • Version diff
    Call centre
    • Call records
    • API logs
    • Manual queue
    • Exception queue
    • Business reports
    • System alerts
    Shared modules
    • Global lineage
    • Operation logs
    • Data permissions
    • Menu permissions
    Offline analysis
    • Real-time landing
    • Standardised loading
    • Decision-process data
    • Variable call logs
    • Main / sub-flow data
    • Experiment data
    Clusters
    • Risk
    • Collections
    • Outreach
    • Marketing
    • Philippines
    • Indonesia
    • Other…
    • Offline cluster
    • Online cluster

    Compute engine & end-to-end integration

    Flink for real time, Spark for batch, one scheduler for both.

    FLINK

    Real-time compute engine
    • Real-time job scaffolding
    • Window aggregation + dimension-table joins
    • Complex-variable latency ≤ 50 ms

    SPARK

    Batch compute engine
    • Offline ETL pipelines
    • Bulk data processing
    • Subject-area data marts

    Unified task scheduler

    UNIFIED TASK SCHEDULER
    1. Flink / Spark task dependency orchestration
    2. Automatic retry on failure
    3. Real-time performance load-tested to target
    4. Four core modules integrated
    5. End-to-end risk flow connected for the first time
    CLOSED LOOP

    End-to-end business loop

    1. Data ingestion
    2. Feature computing
    3. Model scoring
    4. Decision output
    5. Monitoring & alerts

    Key technical breakthroughs

    Five milestones that made the platform work at scale.

    01

    Metadata-driven full-chain processing

    More flexible and maintainable processing of complex data indicators.

    02

    Flink real-time optimisation

    Bloom filters + state-backend tuning bring complex-variable latency to ≤ 50 ms.

    03

    Visual decision-flow editing

    20+ in-house node types; strategy configuration and release are markedly faster.

    04

    Full-chain lineage

    Features, models and decision flows are all traceable — stronger audit, backtracking and regulatory fit.

    05

    Model data-quality monitoring

    Real-time monitoring of model data quality, with minute-level alerts that pinpoint anomalies.

    Put your credit strategy on one risk platform.

    Bring a real scenario — admission, anti-fraud, limits or collections. We'll walk through how it runs on the decision engine, features and models, in your cloud or on-premise.

    What the demo covers

    1. Build a decision flow visually and test it on samples
    2. Onboard a feature and trace its lineage
    3. Canary-release a model and watch PSI / AUC
    4. Deployment, integration and migration plan
    Book a risk demo Talk to an engineer