From core-banking data to a board-ready report — in a few clicks.
koop is an AI-powered data-intelligence showcase built for Co-opbank Pertama (CBP). It demonstrates a Recovery Operations dashboard — powered by AI — that sits on top of CBP’s existing core banking system (Microlink Digital Platform) and deliver what that system doesn’t: flexible, real-time, on-demand reporting.
The gap
CBP runs Microlink Digital Platform — an enterprise core banking system built on Java/J2EE and Oracle. It handles accounts, transactions, and the general ledger reliably. But it does not provide flexible analytics, on-demand reporting, trend prediction, or AI-driven insights. Producing a management report today means manual data extraction, spreadsheet manipulation, and hours of analyst time.
The solution
koop sits on top of the same data as Microlink — via a non-disruptive data extract — and delivers what the core system was never built to provide: instant KPIs, AI-generated board reports, risk-tier analysis mapped to IFRS 9 staging, and trend forecasts with confidence bands. All in a few clicks. No changes to the core system, no data migration, no vendor lock-in.
The USP
Select a product and report type, click generate, and within seconds a structured, board-ready report streams onto the screen — grounded in live computed analytics, not templated filler. That’s the headline capability.
Data fetched and computed live from the database — no stale snapshots, no batch delays.
DeepSeek generates narratives grounded in computed KPIs, risk models, and forecasts. Provider-agnostic — swap to Claude or Gemini via config.
All 4 Syariah-compliant products analysed side by side: Personal, Housing, Commercial, and Revolving Credit.
7-step data pipeline
From raw connection to predictive trends — every step is explorable across 4 layers(Data/Info → Process → Result → Presentation) and 2 viewing modes.
Syariah-compliant products
Unsecured Personal Loan
Secured Housing Loan
Commercial Loan
Revolving Credit Loan
Capabilities at a glance
| Capability | What it delivers | Powered by |
|---|---|---|
| Live AI reports | Streamed Markdown report in seconds — executive summary, portfolio, or credit risk | DeepSeek + computed analytics |
| Default prediction (PD) | Per-facility probability of default + driver explanations | Logistic scorecard + Gradient Boosting |
| IFRS 9 risk staging | 3-tier risk: Stage 1 Performing / Stage 2 Watch / Stage 3 Impaired | PD banding + arrears override |
| Trend forecasting | 6-month projection with 80% confidence bands | Holt-Winters (statsmodels) |
| Ask the Data (NL to SQL) | Plain English questions → safe SQL → AI summary | DeepSeek + read-only query guard |
| Data-quality audit | Live anomaly detection: completeness, duplicates, reconciliation | Python / DuckDB checks |
| Visual analytics | KPI tiles, trend charts, product-mix donuts, risk distributions | Recharts + Tailwind |
| SSR landing and navigation | Matrix-first UX: 7 steps x 4 products x 4 layers x 2 modes | Vike SSR + React |
Non-disruptive
koop reads from a data extract — it never writes to the core banking system. No changes to Microlink, no migration, no downtime. The DataSourceinterface means switching from our synthetic PostgreSQL to CBP’s real Oracle/Microlink data is a config change, not a rewrite.
Provider-agnostic AI
The AI layer uses the OpenAI-compatible protocol and defaults to DeepSeek. Swap the provider, model, or API endpoint via .env— Claude, Gemini, GPT — without touching any code.
Morphable architecture
Two seams guard the architecture: the DataSource interface for the database layer and a typed analytics contract for the Python engine. The analytics engine can evolve from DuckDB → Polars → native C++/Rust without touching the UI.
Synthetic sandbox
All data is synthetic and deterministic — repeatable demos with zero risk of exposing real customer data. The same pipeline runs identically against real data when ready.
See it in action
Explore the matrix, run a live report, or ask the data a question.