TG Intranet · Tesla Group
Corporate intranet that replaced a SaaS CRM and chains of spreadsheets with one system for sales, quoting, people, logistics and reporting.

- Role
- IT lead. Sole developer, project manager and Scrum master
- Period
- Jun 2026 — Present
- Status
- In progress
- Stack
- Kotlin
- Spring Boot
- Angular
- PostgreSQL
- Python
- Redis
- Docker
Tesla Group is an electrical engineering and industrial integration company in Trujillo, Peru. I build and run its intranet as the only developer: an Angular app, a Kotlin core service and a Python data service on one PostgreSQL database. The code and the system are private, so this page describes the architecture and leaves out business data.
What it covers
- CRM: accounts, opportunity pipeline and proposals.
- Quoting with revisions and Excel export, and a price catalog.
- Field rosters and hours, and logistics product coding.
- Sales and project cost dashboards.
- Document comparison: it reads quotation and purchase-order PDFs and proposes fields for a person to verify.
Accounting and sales-order modules exist in rehearsal mode. They mirror the ERP until a cutover.
How it works

- Modular monolith, hexagonal per context. Each business context is a pair of Gradle modules: a core with pure Kotlin domain code, and an adapter module with Spring. Domain modules do not have Spring on the classpath, so importing it does not compile.
- Rules as tests. More than 50 ArchUnit rules cover dependency cycles, isolation between contexts, “every endpoint requires a permission” and “every request body is validated”.
- Integration through the database. The Python service replicates the SaaS ERP through its API into PostgreSQL and rebuilds the reporting tables. There is no HTTP between services. PostgreSQL LISTEN/NOTIFY feeds a WebSocket hub that refreshes open screens.
- Access control. Roles by section and action, with per-user overrides and data scope. Objects outside a user’s scope answer 404. The audit log is append-only, enforced by a database trigger.
- Document processing. A worker with a restricted database role claims runs from an inbox table, extracts with OCR and a language model, and stores a literal citation for each field. It proposes; a person decides.
Engineering practice
- CI pins actions by commit and images by digest, scans with gitleaks, Semgrep and Trivy, and publishes images only from the main branch.
- Integration tests run against real PostgreSQL and Redis. Golden tests reproduce a real quoting workbook cell by cell.
- A sandbox mode runs the same image with synthetic data.
The intranet is also the subject of my capstone project and thesis at Universidad Privada Antenor Orrego.