Private-credit data & analytics
I worked across the data platform, from understanding a lending agreement to delivering the reports investment and operations teams used.
The situation
Daily reporting had grown around SQL queries, Excel, and tightly coupled scripts. A run took more than five hours, and historical reruns and failures were difficult to investigate.
What I did
I mapped the workflow and questioned why each step existed. I removed unnecessary processing, built reusable Python transformations, and shared Parquet datasets across calculations. I ran the new workflow alongside the existing one and reconciled outputs before moving components into production.
What changed
Within my first two months, I brought runtime down to roughly 15 minutes. The same redesign made the pipeline easier to debug, rerun for past dates, and extend.
What I took forward
Before making a process faster, I ask whether it should work that way at all. For financial outputs, I validate the replacement against the existing workflow.
Selected contributions
- Translated legal agreements and term sheets into reusable borrowing-base, covenant, and waterfall calculations; validated definitions with investment, credit, accounting, origination, and operations teams.
- Built ingestion and validation for APIs, CSV, Excel, and file-based data, with reusable FastAPI/Pydantic services, structured errors, audit trails, and standardised outputs for counterparty onboarding.
- Built a bank balance and transaction platform integrating Plaid and MX, with user authentication without credential sharing, provider abstractions, standardised account models, and manual reconciliation.
- Automated FX, SOFR, policy-rate, and central-bank-rate ingestion from financial platforms, APIs, and web sources for valuation and portfolio calculations.
- Built an fsspec-compatible SharePoint access library and SharePoint/Azure ingestion and delivery, including report uploads, archiving, configurable as-of dates, and Teams/email alerts.
- Worked across MCA, education loans, device financing, auto loans, consumer instalment loans, and buy/sell pools, covering onboarding, settlement, servicer normalisation, cohort stratification, and vintage/issuer comparisons.
- Implemented DPD, delinquency roll rates, LTV, concentration and tenor-adjusted exposure, maturity buckets, receivables, margin held, covenant exceptions, collateral performance, and cash-flow monitoring.
- Built rules-driven alerts and reconciliation across loan tapes, payments, schedules, purchase/soldback data, servicers, trustees, and internal systems to catch missing files, duplicates, negative balances, and historical mismatches.
- Investigated a borrowing-base discrepancy in an older deal, traced missing data to silent truncation by a third-party service, regenerated affected reports, and worked with its technology team on remediation.
- Produced scenario and sensitivity analysis for exposure, covenant pressure, credit deterioration, liquidity, and concentration; delivered analytical notes, Excel reports, PowerPoint decks, and executive briefings.
- Standardised time series, date conventions, source mappings, cohorts, assumptions, and control logic so analytical data could support forecasting.
- Built a Databricks proof of concept and ran it alongside the daily production pipeline to evaluate a migration and compare reporting outputs.
- Added chunked loaders, vectorised transforms, schema validation, progress indicators, structured logs, and failure notifications; diagnosed data-load, API, and integration failures using logs and source-level debugging.
- Added CI checks for tests, formatting, linting, types, security, documentation, and deployment validation; containerised workflows and supported rollback and secrets-management controls.
- Configured AI-assisted development and repository guidance for prototypes, implementation, tests, and review while retaining responsibility for architecture, security, and edge cases.
- Documented architecture, data lineage, controls, model assumptions, access patterns, runbooks, and onboarding; supported walkthroughs, knowledge transfer, and audit-readiness reviews.
- Built a Jira-connected hiring workflow for candidate emails, interview booking, and follow-ups at workflow transitions.
Tools and methods: Python, SQL, Pandas, Parquet, Delta Lake, Azure, FastAPI, Pydantic, Docker, Databricks
Have a related challenge?
Discuss a project