AI for Data Analysis¶

LLMs and AI tools accelerate analysis but don't replace judgment.

Pages¶

• Tools — JuliusAI, Claude, Genie, Data Cards
• Prompt Framework — task → context → references → evaluate → iterate

What AI is good for¶

• Data cleaning — generate code to remove nulls, normalize, deduplicate
• Data organizing — pivot, melt, group via natural language
• Documentation — turn code into prose; explain queries
• Reporting — summarize findings into stakeholder-ready text
• Debugging — Python/R/SQL error explanation
• Code generation — boilerplate, transformations, joins
• Synthetic data — mock samples for testing

What AI is bad for¶

• Domain-specific judgment
• Sensitive data outside controlled environments
• Final decisions
• Anything where you can't verify the answer

Privacy and safety¶

• Never paste real PII, financial data, or proprietary code into public LLMs
• Use enterprise tiers with data agreements
• Prefer local models (Ollama, llama.cpp) for sensitive work
• Redact before sending — replace names, IDs, emails

References¶

• Anthropic — Claude
• OpenAI — ChatGPT for Data
• Google — Data Cards Playbook
• Databricks — DASF 2.0
• Ollama — local LLMs