Cognitive Creations Strategy · Governance · PMO · Agentic AI

Agentic AI Engineering Lab — From Prototype to Production

Independent, hands-on engineering lab

Download as PDF

1 — Pitch (short)

Pitch (short)

Pitch (short)

Most trainings explain agents conceptually. This lab is different: it’s a build-focused engineering experience. Participants ship a working RAG agent, evolve it into a multi-agent system with a coordinator, and explore MCP integration with a practical demo — all with evaluation and operational controls.

What makes it different
  • Engineering-first: tools, schemas, validation, and observability
  • Production mindset: evaluation sets, regressions, thresholds
  • Scope mastery: deterministic vs agentic boundaries (“the sweet spot”)
  • Reusable artifacts: starter repos, tool catalogs, and runbooks
60–90 second script
“This lab closes the gap between understanding agents and shipping reliable agent systems. Over the workshop, participants build a grounded RAG agent, then upgrade to a multi-agent coordinator + specialists, and learn how to control quality with tool contracts, deterministic validation, and evaluation sets. We include a practical MCP integration demo and a real-world case (timesheet reconciliation) to teach when logic should be deterministic and when it should be agentic. The goal is reusable systems — not slide-only learning.”
2 — Curriculum (build progression)

Curriculum (build progression)

Curriculum (build progression)

The lab is structured as a progressive build: foundations → architectures → tool contracts → RAG → multi-agent → MCP demo → evaluation/controls → deployment readiness.

Module Topics Hands-on outcome
1) Foundations Agents, models, tokens, cost drivers, constraints, failure modes. Decision map: when to use deterministic logic vs agentic flow.
2) Architectures Patterns: tool-using, RAG+tools, planner-executor, supervisor-specialists. Architecture selector + trade-offs (risk/cost/latency/quality).
3) Tool Contracts JSON schemas, allowlists, validation, tool safety boundaries. Tool catalog + validators (baseline for reliability).
4) Build #1 — RAG Ingestion, chunking, embeddings, retrieval, grounding, evidence outputs. Working RAG agent with evidence + basic tests.
5) Build #2 — Multi-Agent Coordinator + specialists: delegation rules, timeouts, stop conditions, synthesis. Orchestrator + 3 specialists with schema-based outputs.
6) MCP Demo MCP overview + practical Claude integration demo. MCP-enabled demo + hardening checklist.
7) Evaluation & Controls Golden sets, metrics (quality/latency/cost), thresholds, regressions. Evaluation pack + “go/no-go” gates.
8) Deployment Readiness Product workflow, rollout plan, monitoring, incident playbook basics. Release blueprint + runbook template.
Capstone case example
Timesheet Reconciler — a hybrid design where deterministic parsing and calculations live outside the LLM, while the agent handles ambiguity, exceptions, explanations, and decision support.
3 — Format & maximum seats

Format & maximum seats

Format & maximum seats

Cohort caps are designed to preserve hands-on quality and feedback cycles.

Track Duration Max seats Includes
Online Lab (Cohort) 4–6 weeks 20 Live sessions, guided labs, templates, community, office hours.
Online Lab (Premium) 4–6 weeks 15 Everything above + structured code reviews + evaluation guidance + guest AWS deep dive.
In-Person Intensive 10–12 days 12 Daily labs, troubleshooting, demo day, deployment blueprint.
Prerequisites (recommended)
  • APIs (HTTP/JSON) and basic Git
  • Comfort reading/editing scripts (Python preferred, not required)
  • General cloud/dev familiarity (not advanced)
Credential language (safe)
Use: Certificate of Completion Completion Badge
Avoid: “Certificate Program” “Executive Program”
4 — Target profiles

Target profiles

Target profiles

Best for technical practitioners and technical leaders who need build outcomes and operational readiness.

Core technical roles
  • Software Engineers (backend / full-stack)
  • AI/ML Engineers (RAG, eval, observability)
  • Data / Analytics Engineers (retrieval pipelines)
  • Solution / Enterprise Architects (systems + controls)
Technical leadership roles
  • Tech Leads / Engineering Managers
  • Platform / AI Product Managers (technical)
  • AI Leads responsible for delivery and risk controls
  • Consultants delivering repeatable accelerators
Not a fit if…
You want a non-technical overview only, or you won’t have time for hands-on labs (the value is in building).
5 — Target profiles

Target profiles

Target profiles

Best for technical practitioners and technical leaders who need build outcomes and operational readiness.

Core technical roles
  • Software Engineers (backend / full-stack)
  • AI/ML Engineers (RAG, eval, observability)
  • Data / Analytics Engineers (retrieval pipelines)
  • Solution / Enterprise Architects (systems + controls)
Technical leadership roles
  • Tech Leads / Engineering Managers
  • Platform / AI Product Managers (technical)
  • AI Leads responsible for delivery and risk controls
  • Consultants delivering repeatable accelerators
Not a fit if…
You want a non-technical overview only, or you won’t have time for hands-on labs (the value is in building).

Rate this article

—

Share your feedback

Optional: send a comment about this article.