Work in progress— changes will come
Ignite26/27
Back to all phases
Phase 03

November–December 2026

Craft

Build is where you become technically strong. You build your first production-ready agent, implement RAG against real data, and establish an eval pipeline that catches regressions before they hit the client. This is the core — everything else in the program frames Build.

Learning objectives

  • Build a production-ready agent from scratch
  • Implement RAG against enterprise data with measurable quality
  • Establish an eval pipeline as part of CI
  • Integrate via MCP against external services

Core topics

  • 01Agent orchestration: planning, tool use, state management
  • 02RAG engineering: chunking, retrieval, re-ranking, hybrid
  • 03Evaluation as a discipline: golden sets, LLM-as-judge, metrics
  • 04MCP integrations
  • 05Human-in-the-loop patterns

Technologies and frameworks

  • LangGraph · Pydantic AI
  • Langfuse · OpenTelemetry
  • Anthropic SDK · OpenAI SDK · Bedrock
  • Vector DBs: pgvector, Pinecone, Weaviate
  • MCP Inspector · custom MCP servers

Sessions in this phase

  • Masterclass: 'how to set up an eval pipeline'
  • Lab weekend (Q2): 'build the first agent with RAG'
  • Dojo on first production challenges
  • Frontier Friday on new agent frameworks

Artefacts Ignitere deliver

  • 01Running agent with tools and HITL approval
  • 02RAG system with eval report
  • 03Eval framework in Git with golden set and automated testing

Courses in this phase

Shared coursesAll Ignitere

  • Agent-orkestrering med LangGraph

    Bygg en agent som planlegger, bruker verktøy og håndterer tilstand.

  • RAG-engineering og hybrid-søk

    Hvordan finne riktig kontekst og levere forankrede svar.

  • Evaluering som disciplin

    Golden sets, LLM-as-judge og hvorfor eval-rigg betyr alt.

Track courses · Ignite Builder

  • Production agent-patterns

    Streaming, retries, circuit breakers — patterns som holder agenter levende i produksjon.

  • MCP-integrasjoner og custom server-design

    Koble agenter til virksomhetsdata og -handlinger med Model Context Protocol.

  • Eval-CI med golden sets i Git

    Regression-test prompts og agenter — bryt bygget hvis kvaliteten faller.

Track courses · Ignite Tuner

  • Evaluation-håndverk: rubrikker og LLM-as-judge

    Design kvalitetsmål som overlever kontakt med virkeligheten.

  • Vector DB trade-offs

    Velg rett vector-database for din data og workload — pgvector, Pinecone, Weaviate.

  • Graph RAG og hybrid retrieval

    Når flatt RAG ikke er nok — slik finner du svar som krever kontekst.

Track courses · Ignite Shaper

  • Prosessredesign og workflow-mapping

    Finn prosessene som lønner seg å automatisere — og de som ikke gjør det.

Cross-cutting threads

  • EU AI Act practice
  • Consultant craft
  • Evaluation
Phase 02Discover
Phase 04Launch