Creditit Backend Contributions

Problem

An invoice-financing platform needs more than CRUD around invoices and facilities. Credit decisions need real financial signals behind them, collections planning benefits from knowing when a payment is likely to land, and bank statements arrive as PDFs that nobody wants to parse by hand. These needed to slot into an existing, much larger codebase without me pretending I built the whole thing.

My role

My workstreams were a weighted credit-scoring service built on financial metrics, an ML-based payment-date forecasting pipeline, and an OCR pipeline for extracting data from bank statements.

Architecture

The Creditit backend is a Flask, SQLAlchemy, and PostgreSQL service with scheduled jobs and various cloud integrations. My modules plug into that: a scoring service that scores customers from binned financial metrics on a schedule, a forecasting path that calls into an ML pipeline for payment timing signals and exposes it through REST endpoints and CSV export, and a statement-extraction pipeline that chunks large PDFs, runs OCR, and maps the results into schemas for a few different banks with some concurrency to keep it fast. I'm keeping this at the conceptual level. The actual weights, thresholds, and formulas stay proprietary.

Technical highlights

• A weighted credit-scoring service over financial metrics, recalculated on a schedule.
• An ML payment-date forecasting pipeline behind REST endpoints and CSV export.
• OCR-based bank-statement extraction with PDF chunking and concurrent processing across multiple bank formats.