Blackrack Labs¶

A self-hosted AI agent platform, running in production: multi-provider LLM routing with fixed workload tiers, a scoped capability broker, self-hosted SSO and secrets, and a hybrid Kubernetes cluster — operating agents for real users around the clock. Built and operated by Felipe Tejada, it is the working evidence behind his AI Platform / Senior Backend engineering profile.

About the engineer Architecture Routing policy Engineering highlights

Platform architecture — consumers, platform services, providers — One platform, three planes: consumers reach models through LiteLLM and capabilities through the broker; identity, secrets and data back everything.

The platform at a glance¶

~50model aliases · one endpoint

4smart routing tiers

20+LLM & capability providers

15+self-hosted services · one SSO

LiteLLMProvider Brokerk3sAWS + ProxmoxLonghornRDS PostgreSQLAuthentikInfisicalTailscaleHermesCodexClaude Code

What this lab proves¶

Every page on this site documents a subsystem that runs 24/7 for real users — personal assistants, voice reminders, task management, and a fleet of multi-tenant agents — not a slideware architecture.

01 / MODEL PLANE

LLM gateway & routing policy ¶

LiteLLM centralizes about 50 aliases behind one API; fixed workload tiers with audited fallback chains make providers interchangeable and outages become fallbacks.

02 / CAPABILITY PLANE

Provider Broker ¶

Scoped, token-brokered access to search, scrape, speech, voice, browser and MCP — tenants never hold an upstream key.

03 / SECURITY MODEL

Isolation by design ¶

Credential isolation with audits, sealed secrets, SSRF guards, signed webhooks and zero-trust networking.

04 / AGENT RUNTIME

A fleet, not a script ¶

Multi-tenant gateways on k3s with canary rollouts, durable sagas and verified postconditions — agents as production software.

05 / SELF-HOSTED SUITE

SaaS replacement stack ¶

Matrix, Vikunja, Firefly, Plane, Langfuse, SearXNG, Crawl4AI — behind one Authentik login, on own hardware.

06 / THE CAREER CASE

Evidence, not screenshots ¶

How the platform maps to AI Platform Engineer and Senior Backend roles — and where the CV is backed by running systems.

How to read this site¶

Recruiters & managers ¶

Engineering highlights condenses the platform into the skills it exercises.

AI engineers ¶

Routing policy, Provider broker and Harness integration show the routing and capability layers.

Platform / backend engineers ¶

Kubernetes, Security model, Secrets, SSO.

Public scope. This is a sanitized public view of a private production platform: internal IPs, hostnames, tenant identifiers, credentials and billing details are intentionally omitted. If you work with Felipe and need the operational runbooks, they live in the private company repository behind the normal access controls.