A full-stack reconciliation platform built to automate the matching of vendor transaction files against core transaction data. The platform provides a guided 4-step vendor onboarding wizard with intelligent column field mapping, configurable subtype detection rules (keyword/prefix/suffix matching with AND conditions), and automated reconciliation scheduling via Cloud Run. Key features include: multi-file reconciliation (transaction + reversal files with cross-file matching), two-pass reconciliation using primary and fallback matching references, balance anchors for running balance tracking, a readjustment approval workflow, EMTL/IBT/card payment subtype support, upload detail drawers with skipped record visibility, a reconciliation health report pivot grid, and role-based access control. Tech stack: React, TypeScript, Tailwind CSS, Vite, Supabase (auth + DB), BigQuery, Python (Cloud Run backend), Google Cloud Storage.

Designed and maintained the dbt transformation layer for core transaction data across Nigeria and DRC (Democratic Republic of Congo) markets. Built and iterated on models covering daily transaction replicas for corporate and personal wallets, EMTL charge extraction, user loan eligibility scoring, savings plan views, pybot (receipt) orders and accounts, transaction segmentation by type, unified loan models, smart save transactions, and AUM (Assets Under Management) models for the DRC market. Also maintained data quality guardrails — including NULLIF guards for division-by-zero, deduplication of loan statuses, and refactoring of weekly source models. Tech stack: dbt, BigQuery, SQL, Python.

Built and maintained real-time change stream pipelines that capture MongoDB document changes and stream them into Google Cloud Spanner and BigQuery. Extended the pipeline to cover the Nigeria staging environment (ng_staging) and wallet transactions. Hardened the pipeline for production reliability by handling invalid UTF-8 bytes in BSON strings with silent replacement logging and resilient iterator restart logic. Tech stack: Python, MongoDB Change Streams, Google Cloud Spanner, BigQuery, Cloud Run.

Improved the account statement generator service to correctly handle non-NGN transactions (USD, CDF, CAD, EUR, GBP) from international markets. Non-NGN transactions arrive without a linked transaction document, previously causing narrations to show as raw type names (e.g. "Fx Tx Dr", "Wallet Topup"). Built structured narration logic for wallet_topup, fx_tx_dr, p2p, and transfer types using available meta fields (sender_name, recipient_name, narration, RRN), with CREDIT/DEBIT-aware name display. Tech stack: Python, Google Cloud Run, Jinja2, WeasyPrint, GCS.