Sentro ERP

The platform

Sentro is a generic, multi-tenant-ready ERP for manufacturers — built as a product, not a one-off custom job. The codebase is deliberately client-agnostic so it can be deployed for multiple manufacturers, with multi-site support as a first-class concept: every transactional record is scoped to a site and automatically isolated at the data layer.

It covers the full operational and financial backbone — authentication and role-based access, procurement, manufacturing, inventory, sales and receivables, treasury, tax, and financial reporting — all posting into a single general-ledger engine.

Modules

• Procurement — item master, purchase requisitions, purchase orders, goods receipt, and AP with enforced three-way matching
• Manufacturing — bills of materials, production orders, and costing
• Inventory — valuation and adjustments
• Sales, accounts receivable, and collections
• Treasury and vouchers, tax and remittance, branches, notifications, and financial reporting
• A chart-of-accounts and general-ledger engine with period close

Financial architecture

The accounting core is built to standards an auditor would recognize:

• Event-driven GL — operational transactions emit events, and the ledger posts journal entries asynchronously
• Append-only ledger — posted entries are never modified; reversals link back to the original entry, and closing a period locks it
• Immutable audit trail — every financial create, update, and delete records actor, timestamp, before/after values, and IP; nothing is ever hard-deleted
• Three-way match — AP cannot post without reconciling the purchase order and goods receipt within tolerance
• Configurable approvals — per module, per role, with monetary thresholds, expressed as data rather than code
• A single policy seam — every authorization decision routes through one service, leaving a clean path to attribute-based access control

Built agentically

The real story is how it was built: solo, on a night-shift cadence, using a governed team of AI sub-agents — backend and frontend specialists, a financial-logic auditor, a schema reviewer, a test writer, and a project manager — coordinated through a nightly playbook with hard quality gates (financial audit, schema, frontend, audit-trail, and general-ledger checks) before any work is accepted.

The result at roughly 30 weeks: over 1,000 commits, ~1,800 TypeScript files, 113 database migrations, 32 backend modules, 130+ web pages, 38 architecture decision records, and more than 5,000 automated tests — concrete, auditable proof of AI-assisted development at real scale.