About this course
What it takes to keep LLM apps healthy after launch. Model gateways and fallback, prompt and model versioning, evaluation pipelines, tracing, guardrails, red-teaming, cost controls and deployment on cloud infrastructure.
What you will be able to do
- Put a gateway with fallback and budgets in front of models
- Version prompts and run evaluations in CI
- Trace requests across agents and tools
- Defend against prompt injection and data leaks
- Deploy and scale LLM services on the cloud
Curriculum
01The LLM production stack
- Reference architecture
- Failure modes
- SLOs
02Gateways & routing
- Fallback
- Rate limits
- Budgets
03Evaluation in CI
- Golden sets
- Judges
- Regression gates
04Observability
- Tracing
- Cost dashboards
- Feedback loops
05Safety
- Guardrails
- PII
- Red-teaming
06Infrastructure
- Terraform
- Containers
- Autoscaling
Tools you will use
Prerequisites
- Has built at least one LLM app
- Docker basics
Projects you will build
5 milestones
Research agent platform with guardrails, gateway, memory and red-teaming
A search → summarise → write → verify pipeline that produces sourced reports, defended by guardrails and continuously attacked by its own red team.
LangGraph · Bedrock Guardrails · LLM gateway Coming soon
5 milestones
Governed accounts-payable agents for a bank
Agents read invoices, score them for fraud, route approvals and pay the safe ones — wrapped in six layers of governance you can audit after the fact.
LangGraph · Keycloak · LiteLLM Coming soon
5 milestones
AWS cost optimiser dashboard with Terraform and Lambda
Find idle instances, oversized volumes and forgotten snapshots across an AWS account — and see what fixing them would save.
Terraform · AWS Lambda · API Gateway Coming soon