12-Hour Crash
Junior AI Systems Engineer • Remote BD • US Hours

Claude Code, Codex, automation platforms & monitoring.

How to show you can build with modern AI tools — and operate what you build.

12-hour crash planBDT 50k-70k / monthRemote BD • US hoursDiagrams + quiz per module
Fast-learner loop (use for every module): 1) Look at the diagram first — get the shape. 2) Read once, fast. 3) Say it out loud in 30 seconds. 4) Take the quiz. Anything you miss, re-read only that part.

Claude Code & OpenAI Codex — how to talk about them

What they are

Agentic coding tools that read a codebase, plan, edit files, run commands/tests and iterate. You direct, review diffs, and verify — you don't blindly accept output.

Your workflow (say this)

1) Give context & constraints 2) Ask for a plan on bigger tasks 3) Review the diff 4) Run tests 5) Iterate on failures 6) Commit small, explainable changes.

Good practiceWhy it matters
Small, scoped promptsEasier to review; fewer surprise changes
Always run tests / reproduce firstAI can be confidently wrong
Keep secrets out of promptsNever paste API keys or customer data
AGENTS.md / CLAUDE.md context filesProject conventions the agent reuses every session
Demo line: “Here's a repo where I used Claude Code to add a webhook handler. I wrote the test first, had it implement, reviewed the diff, and caught [X] in review.” — bring one real example.

Automation platforms

Tool typeExamplesChoose it when
No/low-code automationn8n, Make, ZapierFast business workflows, many SaaS connectors
Code-basedPython/Node + queues + cronComplex logic, testing, version control, scale
AI workflow / agent frameworksLangChain-style chains, provider SDKsMulti-step LLM logic, RAG, tool calling

Monitoring & reliability checklist

Researching a new AI tool (their requirement!)

  1. What problem does it solve for us? 2) Docs + pricing + limits 3) Small spike (half-day) 4) Eval on our golden set 5) Risks: lock-in, data privacy, reliability 6) Recommendation with trade-offs.

Documentation & SOPs

SOP skeleton

Purpose → Owner → Trigger → Steps → Inputs/Outputs → Failure modes & fixes → Escalation → Changelog. Write docs a teammate can follow at midnight without you.

Build a portfolio project →

flowchart TD Y["You: scoped task
+ context + constraints"] --> A["AI agent
Claude Code / Codex"] A --> P["Plan (big tasks)"] P --> D["Diff / code changes"] D --> R{"You review
the diff"} R -->|Issues| A R -->|Looks right| T["Run tests /
reproduce bug first"] T -->|Fail| A T -->|Pass| C["Small commit
you can explain"]
Figure 8 — The agent loop you should describe in the interview. You direct, review, verify — the agent does not ship alone.
flowchart LR TR["Trigger
form / webhook / schedule"] --> AU["Automation
n8n / Make / code"] AU --> AI["LLM step
classify / extract"] AI --> ST["Store
DB / Sheet"] ST --> N["Notify
Slack / Email"] AU -.-> M["Logs + alerts
failures / cost"] AI -.-> M
Figure 9 — A typical automation from the job description. Every box is something you monitor.

Test yourself

Module 3 Test — AI Tools, Automation & Monitoring

Tap an answer — explanation appears instantly. Pass mark: 70%. Retake until you pass.

Claude Code produces wrong code. Your move?

Why: You own quality: test first, feed back failures, review every diff.

Fast business workflow with many SaaS connectors. Best first choice?

Why: Low-code for speed/connectors; code for complex, testable logic.

Good monitoring for an automation means…

Why: Monitor outcomes (failures, latency, cost), not just uptime.

Before adopting a new AI tool you should…

Why: Spike → eval → trade-off recommendation. That's the research requirement in the posting.

Context files like CLAUDE.md / AGENTS.md are for…

Why: They give the agent durable project context — never secrets.

0 / 5 correct