Agentic Analytics: Build an AI Analyst You Can Trust
Shane Butler
Principal Data Scientist @ Ontra
Sravya Madipalli
Data Science Leader |Ex-Microsoft
.png&w=768&q=75)
This course is popular
12 people enrolled last week.
Build AI Analysts that do the work and prove the numbers are reliable. No code.
Anyone can get AI to produce analysis now. The hard part is knowing whether it is right.
This is a five week, hands on course where you build an agentic analytics system in Claude Code, and then do the thing almost nobody teaches: prove it works.
Weeks 1 and 2 you build. You set up your own environment, write your first skill, design and build a multi agent system, integrate it into a production repo, and connect it to your team's tools through MCP.
Weeks 3 and 4 you make it trustworthy. You measure reliability across repeated runs, triangulate a question across independent methods of analysis, build ground truth, align an LLM judge to human labels, trace every answer back to the query that produced it, and engineer the context that makes the system correct on your company's metrics.
Week 5 you put it to work. You swap the model underneath it, cross check one model against another, and turn the system into value your company can see.
Ten live sessions, two hours each, with three instructors in the room. Every recording is yours to keep. Free course retake. Private AI Builders alumni community. No coding background required, no Python or SQL is ever written by you.
What you’ll learn
Leave with a working agentic analytics system on your machine, a polished analysis, and the patterns and frameworks to make it all yours.
Write skills that persist across context clears, encoding your analytical standards as system behavior rather than conversation.
Run the full build workflow on a real brief: explore, spec, plan, build, test, iterate.
Climb the system ladder, decomposing a single agent into a chained multi agent pipeline.
Survey a developed agentic repo, diff it against your own build, and merge your work in by judgment rather than copy and paste.
Point the analyst at a live warehouse and connect your team's tools over MCP.
Measure reliability across repeated runs and read the variance. An unstable number is an undefined metric, not a model failure.
Instrument provenance so every answer traces back to the query, source, and freshness behind it.
Build ground truth, score against a held out golden set, and cluster failures into ranked modes.
Write the rubric, run the judge blind, and align it to a human labeled golden set before you rely on it.
Read precision and recall on your own judge, isolate the false passes, and tighten the rubric.
Place each kind of knowledge in its right home: resident context, skills, meaning contracts, semantic layer, verified examples.
Write a meaning contract for a contested metric and make independent runs converge on one definition.
Ship a context change with score evidence, reading the per case flip list beneath the aggregate.
Point Claude Code at an open weight model without changing your repo, your skills, or your workflow.
Cross check one model family against another such as Codex, then route the disagreement to act, investigate, or escalate.
Learn directly from expert instructors
.png&w=384&q=75)
Shane Butler
Principal Data Scientist @ Ontra | Ex-Stripe, Nextdoor, PwC, Appfolio
Sravya Madipalli
Data Science Leader@Superhuman (Prev. Grammarly)| Ex-Microsoft, eBay, Nextdoor
.png&w=384&q=75)
Hai Guan
Head of Data @ Ontra | Ex-Nextdoor, LinkedIn, Pinterest, Meta
Who this course is for
Product managers who are tired of waiting in the data team's queue. Stop waiting. Analyze it yourself.
Data scientists who want to multiply their output. You know how to do the analysis manually. Now automate it.
Engineers who want to make data-driven decisions without hiring a data team.
What's included
Live sessions
Learn directly from your instructors in a real-time, interactive format.
20 hours of live instruction with three instructors
Ten sessions over five weeks, two hours each. One instructor teaches, two troubleshoot with you live so nobody gets stuck. Every session is hands on. You are building, not watching.
Weekly live office hours
5 office hours throughout the program to answer general questions and work through the weeks exercises.
Your complete AI analyst repository
A working system you built: a CLAUDE.md that turns Claude Code into your analyst, the skills and agents you wrote, connected data sources, and completed analyses with charts and narrative. All plain markdown you can read and edit.
An eval suite and a calibrated judge
Not just a system, but the machinery to prove it works: ground truth, a held out golden set, an accuracy score, and an LLM judge you built and aligned to human labels.
Recordings, permanently
Every session lands in your inbox the same day and stays yours to keep.
Step by step written guides for every session
Follow along at your own pace, revisit anything you missed, and reference the setup and build patterns later.
Alumni community
Join the alumni channel. Ask questions, share what you have built, and stay connected as the tools change.
Maven Guarantee
Your purchase is backed by the Maven Guarantee.
Course syllabus
16 live sessions • 48 lessons • 5 projects
Week 1
Sep
8
Welcome to Agentic Analytics
Day 1: Welcome to Agentic Analytics
Sep
9
Optional: Install & Setup Office Hours
Sep
10
Optional: Install & Setup Office Hours
Sep
11
Design & Build your System
Day 2: Design & Build Your System
Free resources
Schedule
Live sessions
20-25 hrs
Tue, Sep 8
2:00 PM—4:00 PM (UTC)
Wed, Sep 9
3:00 PM—4:00 PM (UTC)
Thu, Sep 10
3:00 PM—4:00 PM (UTC)
Frequently asked questions
Maven for Teams
Reimbursement
Get your company to pay
Everything L&D needs: email template, receipts, and certificate of completion.
Get reimbursedTeam discount
Learn with your teammates
Save 20%+ when 2 or more teammates enroll in the same cohort.
Save 20%+ with a teamPrivate cohort
Run a cohort for your org
A dedicated cohort with a custom schedule and curriculum, tailored to your team.
Book a private cohort$2,500
USD