Finantrac — Financial Operations Platform
Multi-tenant financial operations platform on ASP.NET Core and PostgreSQL with Plaid bank sync, reconciliation workflows, and ML-assisted transaction categorization.
Overview
Finantrac is a multi-tenant financial operations platform built on ASP.NET Core, C#, PostgreSQL, and Entity Framework Core. It is the most backend-intensive system in my current stack — designed for reconciliation, transaction categorization, and auditability across tenants in live financial workflows.
Key Features
-
Multi-Tenant Accounting Model
Normalized schemas supporting reconciliation, transaction categorization, and full audit trails per tenant. -
Plaid Bank Integration
Secure bank sync and automated transaction ingestion with tenant-scoped data isolation. -
Intelligent Categorization
Fuzzy matching and gradient boosting to classify and route transactions with less manual overhead. -
Platform Foundations
JWT authentication, RBAC, background jobs, response caching, and Swagger/OpenAPI documentation across the API surface.
Technical Implementation
- Backend: ASP.NET Core, C#, Entity Framework Core, REST APIs with structured DTOs and service-layer boundaries.
- Data: PostgreSQL with schema design focused on accounting correctness, tenant isolation, and queryable audit history.
- Integrations: Plaid for financial institution connectivity; background processing for ingestion and categorization pipelines.
- Security: JWT auth, role-based access control, and API documentation via Swagger for internal and partner consumption.
Impact
Finantrac extends production-grade backend work from operations ERP into financial systems — emphasizing schema rigor, integration reliability, and backend ownership from data model through deployment.
Engineering Focus
- Tenant-safe data modeling: Designed schemas and access patterns so reconciliation and categorization stay isolated per tenant.
- Ingestion at scale: Combined Plaid sync with background jobs so bank data lands consistently without blocking API requests.
- Classification pipeline: Blended rule-based fuzzy matching with gradient boosting where manual labels were available.