Codvo Internalby Codvo AI Engineering & Architecture Lead Team

Codvo Internal L2 — Agent Systems Builder

Working controlled agent prototype with tools, data, approvals, observability.

Visit course

Cost

Free

Duration

—

Type

internal_level

Levels

L2 Builder

Target 4wPass ≥ 80%

Rubric (v1)

Criterion	Weight	Must pass
End-to-end functionality	20%	—
Architecture and runtime design	15%	—
Tool and data controls	20%	●
Observability and evaluation	20%	●
Security and failure handling	15%	●
Technical documentation	10%	—

Course material

Full curriculum, deliverables, capstone, and pass criteria.

Duration

4 to 6 weeks, with weekly labs and a final prototype review

Audience

AI engineers, backend engineers, data engineers, QA automation engineers, DevOps engineers, hands-on technical leads.

Prerequisite

Level 1 certification (or equivalent internal waiver from the review board)

Purpose

Certifies that a practitioner can build a working controlled agent prototype and demonstrate practical engineering discipline around tools, data, approvals, evaluation, and observability.

Competencies — practitioners must be able to

Design agent runtime patterns: stateless, workflow-based, or durable
Implement tool calling with constrained permissions
Integrate at least one governed data source or curated knowledge base
Apply basic retrieval quality checks
Add human approval before sensitive actions
Capture traces for model calls, retrieval, tool calls, and decisions
Create a practical evaluation set
Test for prompt injection, wrong tool use, and unsafe outputs
Implement a cancellation, rollback, or kill-switch mechanism
Document production gaps and hardening requirements

Learning modules

1Agent runtime and workflow orchestration
2Tool integration and MCP-style patterns
3Governed retrieval and semantic data access
4Identity and delegated authorization basics
5Human-in-the-loop workflow design
6Observability, traces, and cost monitoring
7Agent testing and evaluation
8Failure handling and rollback

Required deliverables

Working prototype
Technical Architecture Document
Agent Flow Diagram
Tool and Data Access Matrix
Evaluation Plan
Test Evidence
Observability Trace Sample
Risk and Control Register
Production Readiness Gap List

Practical capstone

Customer Support Resolution Agent — workflow: read ticket → retrieve KB content → review customer history/policy data → draft response → human approval → log all calls/retrievals → capture feedback (accepted/edited/rejected).

Assessment method

Demo, artifact review, technical Q&A. Weighted: functionality (20%), architecture (15%), tool/data controls (20%), observability (20%), security/failure handling (15%), documentation (10%).

Pass criteria

Score ≥80% overall. MUST demonstrate working observability, constrained tool access, and one approval/interrupt pattern. A prototype that works but cannot be inspected or controlled does NOT pass.