Advanced AI Analytics Bootcamp
Shane Butler
Principal Data Scientist @ Ontra
Sravya Madipalli
Data Science Leader |Ex-Microsoft
Build production AI analyst system with automated context, multi-agent pipeline
You learned to use Claude Code for analysis. Now build the system behind it. This hands-on weekend bootcamp (4 hours Saturday, 4 hours Sunday) takes you under the hood of a production AI Analyst system with 45+ skills, 25+ agents, and multi-agent pipelines, and teaches you to build your own version from scratch.
The problem every Claude Code user hits: how do you trust the output? A polished analysis can be completely wrong. Inflated numbers from a bad join, causal claims on observational data, a metric definition that changed two months ago. You won't know until it's too late.
This bootcamp solves that. You'll build a multi-agent pipeline where a fresh-context Validator independently reviews every analysis. You'll add multi-layered context with freshness tracking so your AI never works with stale data. You'll wire Karpathy's auto-research loop so your system generates and validates its own context. And you'll add rules and hooks that enforce analytical rigor at the system level.
Everything connects. Each module's output feeds the next. By Sunday evening, you take home two working repos (Claude Code + Codex), a complete AI analyst system you built yourself, and 10 reference guides.
What you’ll learn
You'll learn to build a Claude Code and Codex AI analyst (45+ Skills and 25+ Agents) system you can trust, scale, and take to work.
See the architecture of a production system with 45+ skills, 25+ agents, and multi-agent pipelines used daily
Understand the 5-layer reliability stack: CLAUDE.md, Skills, Rules, Hooks, Subagents, and when to use each
Build your own version from scratch across both days (skills, agents, rules, hooks, context files) and take it to work
Multi-layered context engineering with freshness tracking and confidence grading, so your AI works across long sessions
Karpathy's auto-research pattern: agents mine data sources, auto-generate context files, score quality after each step
Your system stays accurate and gets better over time instead of going stale
A fresh-context Validator that independently reviews every analysis with no shared history or inherited bias
Catches errors a single-prompt analysis misses: arithmetic inconsistencies, hidden composition effects, causal claims on observational
We finally have room to teach validation properly: full instruction, hands-on exercises, real error detection
Warehouse connections via MCP. Connect Claude Code directly to your data without copy-pasting query results
CLI automation: headless mode, model selection per call, tool restrictions, piping. Run Claude Code from scripts
Build your own CLI pipelines using Claude Code and Codex
Rules vs hooks: when to use advisory guidance (~80% adherence) vs deterministic enforcement (100%, no exceptions)
V1→V2→V3 iteration. Never ship the first draft. The 10 extra minutes produce dramatically better output
The 80/20 split: AI handles repeatable analysis, you provide the irreplaceable judgment
Take home your Claude Code AI Analyst system (CLAUDE.md, skills, agents, rules, hooks, auto-research loop)
Take home a Codex-compatible version in AGENTS.md format. Same system, runs in both environments
Open source models in the flow, plus 10 reference guides covering everything from session best practices to production Python scripts
Learn directly from expert instructors
Shane Butler
Principal Data Scientist @ Ontra | Ex-Stripe, Nextdoor, PwC, Appfolio
Sravya Madipalli
Data Science Leader@Superhuman (Prev. Grammarly)| Ex-Microsoft, eBay, Nextdoor
Hai Guan
Head of Data @ Ontra | Ex-Nextdoor, LinkedIn, Pinterest, Meta
Who this course is for
Product managers already using Claude Code. You've built skills and agents. Now build a system that validates its own output.
Data scientists who know how to work with Claude Code. Now automate the full workflow: context loading, analysis, validation, and narrative.
Analysts and engineers comfortable with skills and agents who want to go deeper in Claude Code.
What's included
Live sessions
Learn directly from your instructors in a real-time, interactive format.
Claude Code AI Analyst + Codex Repos(45+Skills and 25+Agents)
A working system. You get a configured CLAUDE.md that turns Claude Code into your analyst.
8 hours of live instruction (Saturday + Sunday, 4 hours each)
Saturday + Sunday, 4 hours each day. One instructor leads. Two troubleshoot and help you in real-time so nobody gets stuck. Every section is hands-on, you are building, not watching. Recordings of both days sent to all participants
Alumni community
Join a community of bootcamp alumni. Ask questions, share findings, get help troubleshooting, and see how others are using their AI analyst repos at their companies. Stay connected as the tools and techniques evolve.
Step-by-step guides for every section
10 reference guides covering best practices, Python scripts, and practitioner workflows
Lifetime access
Go back to course content and recordings whenever you need to.
Certificate of completion
Share your new skills with your employer or on LinkedIn.
Maven Guarantee
Your purchase is backed by the Maven Guarantee.
Course syllabus
Week 1
Jun
6
Bootcamp Day 1
Lesson 1: The Validation Failure Demo
Lesson 2: AI Analyst under the hood (Skills + CLAUDE.md + MCP Connections)
Lesson 3: Context Quality + Freshness Tracking
Lesson 4: Build the 5-Agent Analytics Pipeline
Lesson 5: Add the Fresh-Context Validator
Jun
7
Bootcamp Day 2
Lesson 6: CLI Automation + Karpathy's Auto-Research Loop
Lesson 7: Rules, Hooks, and the 5-Layer Audit
Lesson 8: Practitioner Patterns and Multi-Cursor Analysis
Lesson 9: Codex Demo + Dual Repo Handoff
Lesson 10: Open Source Models + Wrap-up
Schedule
Live sessions
8 hrs
Sat, Jun 6
2:00 PM—6:00 PM (UTC)
Sun, Jun 7
2:00 PM—6:00 PM (UTC)
Frequently asked questions
$1,500
USD