AI platform engineer (LLMOps)
Run AI in production: cost, uptime and safety, at scale.
At the end: Put models behind a gateway with budgets and fallbacks, measure them, secure them and run the stack on containers and Kubernetes.
Stage 1: Foundations
What the model API costs and where it breaks.
- Python, properly: The Python the platform code is written in.
- Building with LLMs: Tokens, streaming, retries and limits on the raw API.
Stage 2: Run AI in production
The platform between your apps and the models.
- Build an AI gateway: One gateway over many providers: fallbacks, caching, budgets and tracing.
- Evals that catch regressions: Know a model change is safe before it ships.
- AI security: attack and defend: Injection, leaks and tool abuse, and the defences against them.
- Fine-tuning basics: When a smaller tuned model beats a bigger one, and how to tell.
Stage 3: The infrastructure
Where the platform runs.
- Docker, start to ship: Package every service the same way.
- Kubernetes: Deployments, services and rollouts.
- Redis: Caches and rate limits in front of the models.
- Linux on call: Debug it at 2 am.
Stage 4: Interview ready
The production incidents interviews ask about.
- AI engineering traps: AI bugs from production: cut-off replies, retries and cost blowups.
- Concurrency and races: Concurrency bugs and how to prove them.