Piyush Barik

All case studies

Ascentium · 2025 to present

Payment Reconciliation & Settlement

AI-assisted reconciliation that matches wire and card payments to invoices, with a full audit trail.

Anonymised. Client names and operational detail withheld.

Role

Full-stack Developer

Year

2025 to present

Stack

  • Next.js 16
  • TypeScript
  • .NET 10
  • .NET Aspire
  • SQL Server
  • OpenAI
  • TanStack Query
  • Recharts
  • Docker
  • Azure Kubernetes
  • Azure DevOps

Overview

I was the primary full-stack owner of an AI-assisted reconciliation workspace. An OpenAI and Azure Document Intelligence pipeline parses invoices into structured line items that drive a matching engine across two payment rails.

Challenge

  1. Reconcile two payment rails, bank wire transfers and card payments, against open invoices.
  2. Parse inconsistent invoice documents into reliable structured line items for matching.
  3. Make confidence-scored matches reviewable, with exceptions handled safely and an immutable audit trail.

Approach

  1. Owned the settlement endpoints, PDF-evidence serving from blob storage, and the manual settlement-routing logic.
  2. Built the payment review drawer, split-payment and manual-override dialogs, and the audit UI.
  3. Wired an OpenAI + Document Intelligence parsing pipeline into the confidence-scored matching engine.

Outcomes

  1. Was the primary full-stack owner across both repositories.
  2. Fixed floating-point rounding in charge allocation and tightened logging hygiene.
  3. Turned a manual reconciliation process into an auto-and-assisted settlement workflow.

Curious about the longer version, or the parts that didn't make the page? I'd be happy to talk it through. Get in touch