01Case study
BrTools and
BrTools AI.
The internal operations platform I created and maintain at BrSuper Internet, an ISP with over 60,000 subscribers. BrTools belongs to BrSuper and its code is private, so this page covers my work on it.
~200
People using it
5
Areas served
1 h to 5 min
Provisioning procedures
0
Downtime in the database switch
02BrTools AI
Operations, from a chat.
An agent built into the platform and used across the whole company.
What it does
Teams open and schedule work orders in the ERP, handle GLPI tickets and diagnose customer connections over TR-069, with lookups in Zabbix and the internal wiki.
How it's built
Each module and integration declares its own tools, which the AI core discovers through NestJS and limits to each user's permissions. Actions that change data ask for confirmation before they run.
03The platform
What it took.
Support, NOC, Operations, Sales and Field all work in it, and each brought different problems.
Architecture
Modular NestJS with DDD, Clean Architecture and CQRS. It's multi-tenant, with an isolated stack per customer: its own application, database, Redis, proxy and Grafana. Modules are pluggable per installation, with 15 in production.
Network automation
Provisioning on network equipment uses one driver per vendor, talking Telnet over SOCKS proxies to the OLTs.
Integrations
The platform connects to the company's ERP over REST and direct database access, covering customers, contracts, invoices and work orders, plus third-party services that only deliver events by webhook.
Live database migration
I led the production switch from MariaDB to PostgreSQL with no downtime. Both generations of the API served traffic for the entire transition.
Fleet management
A full domain, from vehicles to the fuel financial cycle, with document issuing, an approval chain and reconciliation at closing.
Observability
OpenTelemetry instrumentation lives in a reusable package, and Grafana dashboards are provisioned as code, with traces, metrics and logs separated by customer.
Operations bot
A conversational engine with its own state machine for the flows and each session in Redis, which brought platform operations into the chat.
Product discovery
I worked with the five areas directly: gathered their pain points, wrote the spec for each feature and prioritized the roadmap from those conversations.