← Role tracks

R1 Stage D · 160 h (48 T / 112 P) Anchor track

AI Engineer

Applies pre-trained models and existing AI tooling to build and operate production AI features; does not train foundation models.

Tech stack

PythonFastAPILangChainLangGraphLlamaIndexpgvectorQdrantLiteLLMRedisDockerRAGASLangSmithGitHub Actions

Modules

M1 · Production RAG engineering

40 h
TOPICS
  • Query understanding and rewriting
  • Multi-hop and recursive retrieval
  • Reranker selection and tuning
  • Chunk and index experimentation methodology
  • Freshness and invalidation
  • Multi-tenant corpus isolation

Lab: Systematic retrieval-tuning study raising recall@k on a fixed corpus with documented method

Course material for this module is in production.

M2 · Agentic feature delivery

40 h
TOPICS
  • Workflow decomposition
  • Tool contract design
  • Durable state
  • Retries and compensating actions
  • Partial failure handling
  • Human-in-the-loop checkpoints
  • Multi-agent supervision

Lab: Multi-step agent with durable state, approval gate and full recovery behaviour

Course material for this module is in production.

M3 · Reliability, cost and latency engineering

40 h
TOPICS
  • Model routing and fallback chains
  • Provider outage handling
  • Semantic and exact caching
  • Streaming
  • Concurrency and backpressure
  • Rate-limit management
  • Token budgeting
  • Unit economics
  • SLOs and error budgets

Lab: Reduce cost per query by ≥50% and p95 latency by ≥30% at held quality

Course material for this module is in production.

M4 · Ship and operate

40 h
TOPICS
  • Service packaging
  • Configuration and secrets
  • CI/CD with evaluation gates that block regression
  • Canary release and rollback
  • Tracing and alerting
  • On-call runbooks
  • Post-incident review

Lab: Deploy behind an eval-gated pipeline; force a regression and prove the gate blocks it

Course material for this module is in production.

Track project

Enterprise knowledge assistant: RAG over a multi-format corpus with enforced citations, an agentic action with approval gate, layered guardrails, evaluation harness separating retrieval and generation, documented cost and latency budget, deployed via CI/CD with eval gates, tracing and a runbook.

Job-ready exit standard

Ships a RAG/agent system end to end and defends design under cost, latency and safety constraints; 4+ repositories; interview-ready on RAG and agent system design, evaluation methodology and inference economics.