CPT ERP System
Core B2B ERP for a large office-supplies distributor operating across multiple sales channels in Argentina. Operational backbone of the business — customers, orders, stock, invoicing, payments, logistics, and real-time external order ingestion under high concurrency.
RoleSenior Backend Engineer / hands-on Tech Lead — architecture, domain design, integrations, performance, and technical mentorship across a team of 5–7 developers.
- 2,000–3,000 daily orders, peaks of 2,000 orders/minute during campaigns
- Integrations with MercadoLibre, Shopify, VTex, Stripe, PayPal, and MercadoPago
- Three embedded intelligence modules: warehouse, procurement, and credit risk
- Picking error rate reduced by ~40% through adaptive warehouse recommendations
- Node.js
- TypeScript
- Express
- TypeORM
- MySQL
- Vue.js
- Quasar
- RabbitMQ
- Docker
- Kubernetes
Technical deep dive
Location Recommender
All orders are assembled manually at scale. The system tracks picking errors grouped by product similarity, such as same supplier or same category, identifying which article combinations generate the highest confusion rates. When a pattern crosses a threshold, the system generates a warehouse rearrangement suggestion. If the operator accepts it, the system enters a follow-up cycle spanning several days, monitoring subsequent picks on those articles until a statistically meaningful volume is reached, then reports whether the error rate improved. This closed feedback loop reduced picking errors by around 40%.
Purchase Recommender
Automated purchasing suggestions for the procurement team based on product rotation rates, supplier lead times, historical sales data, and seasonal demand patterns. When an operator approves a suggestion, the system automatically generates a purchase order and dispatches it to the supplier. Supplier delivery performance is then tracked against committed lead times; if a supplier consistently delivers early or late, the system adjusts future recommendation parameters accordingly. This created a self-improving procurement loop that became more accurate over time as supplier behavior data accumulated.
Credit Risk Recommender
Assisted order approval workflow built around each customer's payment history, outstanding orders, current balance, and post-dated checks in portfolio. The system automatically pre-approves or pre-rejects incoming orders, leaving the sector manager to manually review only the borderline cases. Risk parameters are fully configurable at multiple levels of granularity: global thresholds apply by default, but can be overridden per salesperson, per customer, or per time window. This layered configuration gave the business fine-grained control over credit exposure without requiring manual review of every order.
View images
