Blog

Suggested pace, in weeks

1. 1. Foundations weeks 1 to 6

   You build: A program that calls one model from Java and Python and logs tokens and cost

2. 2. LLM apps and RAG weeks 7 to 14

   You build: A document Q&A service in Spring Boot that shows its sources

3. 3. Agents and MCP weeks 15 to 22

   You build: A ticket agent that drafts replies and waits for approval

4. 4. Production weeks 23 to 28

   You build: Evals and tracing for your agent, plus a short write-up

Runtime policy check

1. 1Agent planned action
2. 2Runtime policy engine: cost and rate limits, roles and scopes
3. 3JSON Schema validator
4. 4Ephemeral execution container, then verified result

Hybrid retrieval

1. 1User goal or sub-task query
2. 2Hybrid search: BM25 keyword matching and vector embeddings
3. 3Re-ranking model (cross-encoder)
4. 4Filtered context window, then the agent loop

orchestrator.ts

1export async function dispatchAgentStep(2  request: AgentRequest,3  governance: RuntimeGovernance4): Promise<DispatchResult> {5  const startTime = Date.now();67  // 1. Evaluate runtime policy8  const policyCheck = await governance.validateIntent(request);9  if (!policyCheck.permitted) {10    throw new Error(`Execution blocked by policy: ${policyCheck.reason}`);11  }1213  // 2. Select model tier based on task complexity14  const targetTier = request.requiresReasoning15    ? ModelTier.REASONING_FRONTIER16    : ModelTier.FAST_UTILITY;