AI Coding Exercises — Data skill for Claude Code
Hands-on practice for AI engineering interviews: build LLM evals from scratch with the Claude API, plus drills on RAG, tool-use agents, validating LLM output, eval metrics, async API clients, data cle.
How to install AI Coding Exercises
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open patrickporchejr/ai-coding-exercises and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What AI Coding Exercises does
Hands-on practice for AI engineering interviews: build LLM evals from scratch with the Claude API, plus drills on RAG, tool-use agents, validating LLM output, eval metrics, async API clients, data cleaning, record matching, error triage, and database contention.
Alternatives in Data
- Vibepod CLI — Unified CLI for running AI coding agents in isolated containers 138 ★
- Exploitation Validator — A prompt-based pipeline for finding, validating, and proving vulnerabilities using LLM sub-agents 94 ★
- Fire Analytics — View skills usage analytics and effectiveness metrics 77 ★
README
AI Coding Exercises
Hands-on practice for AI engineering interviews. Each exercise is written like a real interview question. You build the solution from scratch, then answer debrief questions the way you would in the room.
**Who it's for:** engineers preparing for interviews that involve LLMs, such as building evals or handling model output, plus the backend and data problems that come up alongside them.
**What's here:**
- An eval track (exercises 1–3): build a prompt-evaluation pipeline in Python, first guided and then timed without AI, and finally prove your results can be trusted.
- Drills (exercises 4–12): standalone problems on reliability, databases, data cleaning, RAG, tool-use agents, and the metrics and validation work that surround LLM systems.
Exercises
Eval track: do these in order
| # | Exercise | Mode | Time box |
|---|---|---|---|
| 1 | Support-ticket triage | Guided: docs, the cheatsheet, and AI help are all allowed. Go slowly and understand each piece. | none (aim for 2–3h) |
| 2 | Text-to-SQL | Timed, no AI: official docs only. No AI assistants, and no copying from exercise 1. | 60 min |
| 3 | Trust your eval: meeting action items | Timed, with AI: the test set and two prompts are provided. Check the LLM judge against your own labels, then decide whether v2 really beats v1. | 60 min |
Read only the prompt you're working on. Don't look at exercises 2 or 3 until you sit down to do them.
Drills: any order
| # | Exercise | Skills | Needs |
|---|---|---|---|
| 4 | Contention: auction bids | Race conditions, row locking, optimistic concurrency, isolation levels, deadlocks | Docker |
| 5 | Clean a messy CSV | Parsing, normalization, edge cases, explaining your choices | P |
Related Skills
Eval From Scratch
Build an LLM-as-a-Judge eval from a pasted system prompt or a description — no observability data needed.
Mysh
A safer MySQL CLI for AI coding agents: connection profiles, SSH tunnels, and automatic sensitive-data masking
DPDP ACT Compliance
Claude Agent Skill for India's DPDP Act 2023 & DPDP Rules 2025. Section-cited legal guidance plus hands-on web
Usc Econ AI Workshop
Materials and code from USC Economics Department AI workshops (Spring 2026). Hands-on sessions on GPT for data
Myrag
Free, local RAG over messy government PDFs and Word docs (NIST AI RMF, CSF 2.0, FedRAMP): hybrid search in Pos
Verifyconversion
Verify Phase 1 markdown conversion quality by validating table structure, data integrity, and completeness
Related Agents
AI Data Specialist
Deep AI/data engineer — LLM integration and agent systems, RAG, evals, data pipelines, and ML productionizatio
AI LLM
AI/LLM section head: builds features that call Claude or other models: prompts, tools, RAG, evals.
ML Engineer
Use for ML/AI model work — training, fine-tuning, evaluation, RAG, agents, embeddings, evals, deployment, MLOp