A selection of data engineering, credit risk, and AI projects delivered for financial institutions and credit providers.
A large financial institution was running month-end reporting across 11 disconnected source systems — core banking, GL, CRM, and loan origination — resulting in a 6-day close cycle, frequent reconciliation failures, and no single trusted view of the balance sheet.
Designed and delivered a centralised cloud data warehouse on Azure Synapse, with automated ELT pipelines ingesting and reconciling data from all 11 source systems nightly. Built a layered data model (raw → staging → conformed → reporting) aligned to BCBS 239 data lineage requirements.
A consumer credit provider was using a legacy scorecard built on 3-year-old bureau data. Approval rates were declining and bad debt was rising — the model no longer reflected current borrower behaviour in a post-COVID credit environment.
Rebuilt the probability of default (PD) model using gradient boosting on 24 months of behavioural, bureau, and transactional data. Included full model documentation, champion/challenger testing infrastructure, and a monitoring dashboard tracking Gini, KS, and PSI monthly.
The finance team at a mid-size bank was manually calculating IFRS 9 expected credit loss (ECL) provisions in Excel each quarter — a 10-day process prone to error, version conflicts, and audit risk. Regulatory submissions were consistently late.
Built an automated ECL calculation engine that stages loans, applies PD/LGD/EAD models per IFRS 9 stage, and produces a full provision report with audit trail. Integrated directly into the GL for journal entry generation and connected to the PA reporting template for regulatory submission.
A bank with five product lines (home loans, personal loans, vehicle finance, credit cards, and transactional accounts) had no unified view of a customer's full relationship. Marketing, credit, and collections teams each operated from different, conflicting customer records.
Designed and built a master customer data platform with entity resolution logic to deduplicate and link records across all five product systems. The resulting SCV feeds the CRM, the credit risk engine, and a real-time customer profitability model used by relationship managers.
A credit provider's collections team was working a flat call list — every account in arrears treated equally. High-value accounts that were likely to cure were receiving the same attention as low-value chronic defaulters, wasting agent time and missing recovery opportunities.
Built a collections scoring model that predicts the probability of cure within 30 days for each arrear account, weighted by exposure at risk. The model feeds a daily prioritised work queue for collections agents, with a management dashboard tracking roll rates, cure rates, and agent performance.
A digital payments provider was running batch fraud detection on a 4-hour lag — by the time alerts fired, fraudulent funds had already moved. The existing rules-only engine was also generating a high volume of false positives that overwhelmed the compliance team.
Deployed a streaming transaction monitoring pipeline processing events in real time, combining a rules engine with an ML anomaly detection layer. The ML layer learns customer spending patterns and flags deviations, dramatically reducing false positives while improving the true fraud catch rate. Alerts route to a case management dashboard for analyst review.
Project details have been anonymised to protect client confidentiality. Outcome figures reflect actual results achieved. Technologies listed are those used in each specific engagement.