Implementation — Banking & Compliance Hub

Concept → Tool

Every Concept Has a Running Implementation

This page is the bridge between theory and practice. Click any concept to see its detailed explanation.

Concept	Tool in Demo	Files	Learn More
Legacy Data Sources	PostgreSQL, MySQL, SQLite (Docker)	docker-compose.yml	Architecture Governance
Mock Banking Data	Python Faker — 1 000 customers, 10 000 txns	mock_data/generate_data.py	Compliance
Data Catalog & Lineage	OpenMetadata (Docker)	openmetadata/connectors/	Architecture Governance
Data Quality Profiling	Great Expectations	great_expectations/	Data Contracts
Data Contracts	YAML + validate_contracts.py	contracts/*.yml	Data Contracts
Semantic Model	dbt — staging + customer_360 mart	dbt_project/	Semantic Layer
Knowledge Graph	Neo4j (Docker) + Python loader	knowledge_graph/load_graph.py	Knowledge Graph
Ontology	RDFLib (Python) — OWL class hierarchy	ontology/banking_ontology.py	Semantic Layer
AI Agent	Ollama (Llama3/Mistral) + Neo4j Cypher	agent/agent.py	AI Agents

Step-by-Step

From Zero to Working Demo

Run these commands in order. Each maps to a concept layer in the architecture.

① Start Infrastructure

Spin up all databases + OpenMetadata + Neo4j:

cd banking-demo
docker compose up -d
# Wait ~60s for all services to be healthy
docker compose ps

② Generate Mock Banking Data

Creates 1 000 customers, 1 000 accounts, 500 loans, 10 000 transactions across all three databases. Idempotent — safe to run twice.

pip install -r requirements.txt
python mock_data/generate_data.py
# Intentional issues injected: 40 null branch_codes, 3 negative balances

③ Run Data Contracts — WOW MOMENT 1

Watch the validator catch the injected quality issues before they reach downstream consumers.

python contracts/validate_contracts.py

# Expected output:
# [FAIL] savings_account — quality_rule: no_nulls on branch_code (40 violations)
# [FAIL] savings_account — quality_rule: value_range on bal_amt (3 negative values)
# [PASS] loan_position — all rules passed

④ Run Great Expectations Profiling — WOW MOMENT 2

Automated profiling discovers distributions, anomalies, and the T-1 inconsistency between PostgreSQL and MySQL balances.

great_expectations checkpoint run banking_checkpoint
# Opens HTML report: distributions, null counts, anomaly flags

⑤ Build Semantic Layer with dbt

Staging models + customer_360 mart with total_exposure, overdue_flag, and balance_category.

cd dbt_project
dbt deps
dbt run
dbt test
cd ..

⑥ Load Knowledge Graph

Loads all customers, accounts, loans, and branches into Neo4j. Creates OWNS, HELD_AT, DISBURSED_AT relationships.

python knowledge_graph/load_graph.py
# Output: 1000 customers, 1500 accounts, 500 loans, N branches loaded
# Visit http://localhost:7474 to browse the graph visually

⑦ Run the AI Agent — WOW MOMENT 3

Natural language questions answered by Ollama-generated Cypher with node-level provenance.

# Make sure Ollama is running: ollama serve
python agent/agent.py

You: What is customer CUST_0042's total financial exposure?

Agent: Generating Cypher...
MATCH (c:Customer {cust_id: 'CUST_0042'})-[:OWNS]->(a)
RETURN c.name, sum(coalesce(a.bal_amt,0) + coalesce(a.outstanding_amt,0)) AS total

Answer: Rajesh Kumar has total exposure of ₹4,82,310
Nodes traversed: Customer[CUST_0042] → SavingsAccount[SAV_0042] → LoanAccount[LON_0021]
Source: Neo4j graph (loaded from PostgreSQL ACCT_MASTER + LOAN_HDR, 2026-05-15)

🌟 The Three "Wow Moments" for a Client Walkthrough

Wow 1 — Contract Catches It"We intentionally injected 40 null branch codes and 3 negative balances. The contract validator catches them in 2 seconds — before they reach the risk report. Without this, they'd silently corrupt the regulatory submission."
Wow 2 — Profiling Finds the Inconsistency"Great Expectations shows the balance distribution gap between PostgreSQL (live) and MySQL (T-1). This 3-system inconsistency is invisible to the naked eye — profiling makes it measurable and alertable."
Wow 3 — Agent Answers with Provenance"A non-technical user asked a natural language question. The agent generated Cypher, ran it against the knowledge graph, and returned the answer with a node-level audit trail. No SQL training required. Fully governed."

Reference

Project File Structure

banking-demo/
  docker-compose.yml          # PostgreSQL + MySQL + Neo4j + OpenMetadata
  .env                        # All secrets (never committed)
  requirements.txt            # Pinned Python dependencies
  mock_data/
    generate_data.py          # Idempotent data generator
  contracts/
    savings_account.yml       # Schema + freshness + quality rules
    loan_position.yml
    validate_contracts.py     # PASS/FAIL validator
  dbt_project/
    models/staging/           # stg_acct_master, stg_loan_hdr, stg_txn_log
    models/mart/
      customer_360.sql        # total_exposure, overdue_flag, balance_category
    metrics/
      portfolio_at_risk.yml   # Semantic metric definition
  great_expectations/
    expectations/             # Per-table expectation suites
    checkpoints/              # banking_checkpoint.yml
  ontology/
    banking_ontology.py       # RDFLib OWL: BankAccount hierarchy + GDPR annotations
  knowledge_graph/
    load_graph.py             # Loads Neo4j from PostgreSQL (idempotent MERGE)
  agent/
    agent.py                  # Ollama CLI agent: NL → Cypher → answer + provenance
  openmetadata/
    connectors/               # Ingestion configs for all 3 databases
  docs/                       # This GitHub Pages site

← AI Agents Home →

Demo Implementation