Pre-launch · Early access

Code is cheap.
Judgment is
what gets you hired.

Klerisy hands you a real project in your domain — data loaded, environment running — and an AI that guides you rather than give you the answer. You steer. You decide. You learn.

Join the Waitlist →How it works ▶
Step 3 of 8  ·  38%
notebook.ipynb
Step 3 — Exploratory Data Analysis
Understand the data before touching a model. Spot problems early.
[1]
import pandas as pd import numpy as np # Load gene expression dataset df = pd.read_csv('gene_expression.csv') print(df.shape) print(df['subtype'].value_counts())
Out [1]
(523, 1842)
subtype
0    441  → 84.3%
1     58  → 11.1%
2     24  → 4.6%
[2]
# Check missingness & feature ranges missing = df.isnull().sum().sum() print(f"Missing: {missing}") summary = df.describe().loc[[     'mean', 'std', 'min', 'max' ]].round(3)
Out [2]
Missing: 0
       gene_0001  gene_0002
mean     0.002     0.114
std      2.841     2.903
min     -4.231   -3.981
max     12.340   11.883
[ ]
# Your next analysis here...
# What should you look at next?
Klerisy Tutor
Gene Expression Classification
Klerisy
Good. 523 samples, 1,842 features. What does that ratio immediately suggest?
You
p ≫ n — curse of dimensionality?
Klerisy
Exactly. What technique addresses that before you model?
Think about what PCA does to the feature space.
You
Dimensionality reduction first — PCA to compress the 1,842?
Klerisy
Right. Now — subtype 0 is 84% of the data. What metric should you not use, and why does it fool most beginners?
↑

Every job posting wants experience. Every course gives you a certificate. Klerisy gives you the work.

The gap

You finished the course.
You still can't start the work.

Courses optimise for completion. Real analysis rewards the ability to sit with an ambiguous dataset and decide what to do first.

What courses train

Clean CSVs with one right answer Copying the instructor's cell Accuracy on a balanced toy set A certificate nobody reads

What the job asks

35% missing and no documentation Choosing between three defensible paths Knowing why your metric is lying Explaining it to someone senior

What Klerisy builds

Judgment under real ambiguity The habit of interrogating your own results Language for defending a decision A repo you can walk someone through

How it works

Zero setup. Real work. Actual thinking.

Every project comes fully loaded — data, environment, and an AI guide that asks questions instead of giving answers.

Step 01
Pick a project in your field

Biology, finance, clinical data, operations. Choose a domain you already know — the data science will feel immediately relevant, not abstract.

No DS experience needed
Step 02
Everything is already set up

Python, Jupyter, every library — running in your browser. No installs, no environment errors, no Stack Overflow rabbit holes. Open and code.

Browser-based · instant start
Step 03
Your AI guide thinks with you

Stuck? It doesn't hand you the answer — it asks you the right question. You understand every line you write and every decision you make.

Socratic AI tutor
Who it's for

Two people. Same missing piece.

01
The new graduate
I know the algorithms.
I freeze the moment
the data gets messy.
Today
Coursework done, maybe a master's — and every posting still asks for experience you were never given a chance to get.
After
Three or four projects you can talk through for twenty minutes without notes, because every call in them was yours.
→ Interview answers, not certificate screenshots
02
The domain expert
I've had years of data
in front of me. I've never
been able to use it.
Today
Engineer, clinician, researcher, analyst. You know the field better than any bootcamp graduate ever will — and none of it is in your analysis.
After
A four-project specialization in your own field, defensible in a review and legible on a résumé.
→ The translation layer, not another intro course
Project library

Every project hides
a trap worth learning.

Not exercises. Each one is built around a specific mistake that senior data scientists have learned to smell before they make it.

Biology · Genomics
Gene Expression & Cancer Subtype Classification
RNA-seq from 500 tumour samples. Find the subtypes — then work out why your first model looked so good.
Imbalanced classesDim reductionAccuracy paradox
Intermediate
8–12 hrs
→
Clinical · Epidemiology
ICU Readmission with 35% Missing Data
Before you model anything you have to decide what the gaps mean. Every choice moves the answer.
MissingnessLeakageData leakage
Advanced
12–18 hrs
→
Finance · Risk
Credit Default When the Past Stops Predicting
The model trained beautifully. Then the regime changed. Learn to see distribution shift before it costs money.
Distribution shiftCalibrationRegime change
Advanced
10–16 hrs
→
Operations · Supply chain
Demand Forecasting in a World of Outliers
Sales history with pandemic-era anomalies baked in. A naive model looks great. Decide what you are actually forecasting.
OutliersTime seriesOutlier bias
Beginner
5–8 hrs
→
Policy · Social science
Causal Inference in Education Interventions
Does the tutoring programme work, or does selection bias make it look that way? Correlation, causation, confounding.
Causal inferenceSelection biasConfounding
Intermediate
8–12 hrs
→
Engineering · Sensors
Predictive Maintenance on Vibration Data
Failures are rare, sensors drift, and labels arrive late. Build something an operations team would actually trust.
Rare eventsSignal driftLabel leakage
Intermediate
8–14 hrs
→

Showing 6 of 24 projects. The rest open at launch.

Join waitlist for the full library →
What you walk away with

Three things, and none
of them is a certificate.

01 · The repo
Pushed to your GitHub

Notebook, commits, and a README written from your own decisions — not a template. It reads like work because it is work.

02 · The intuition
Knowing when to distrust yourself

Leakage, shift, imbalance, autocorrelation — you meet each one as a problem that bit you, which is the only way it sticks.

03 · The story
Answers for the hard interview

"Why that model?" "What would you do differently?" You wrote the reflection at step eight. You already have the answer.

Pricing

Pay for what you
actually want to learn.

Buy one project, take a whole specialization, or bring Klerisy to a team. One-time payments — no subscription unless you want one, and waitlist members lock early-access pricing.

Per project
Single project
Buy exactly the project you want. Lifetime access, full tutor, GitHub push. No commitment.
$49
one-time · lifetime access
✓Full notebook, all 8 steps
✓Unlimited Socratic AI tutor
✓GitHub repo created and synced
✓Auto-generated README
✓Draft Medium & LinkedIn posts written from your own notes
Join waitlist
Most popular
Specialization
Track
Four to six projects sequenced so each one assumes the last. Cheaper than buying them separately.
$179
one-time · saves ~$65
✦Everything in Single project
✦4–6 projects in one learning arc
✦Tutor context carries across projects
✦Specialization certificate
Join waitlist
Teams
Custom
per seat · annual billing
For research labs, data teams, and graduate programs.
✓Everything in Full Access
✓Admin dashboard & cohort tracking
✓Custom domain-specific project creation
✓Team progress reporting
✓Dedicated onboarding call
✓SLA & invoiced billing available
Get in touch →
Premium — all access
Everything, unlocked.
Every specialization, every project, plus cohorts with peer review and live sessions. Less than two tracks bought separately.
✦All tracks, all 24+ projects
✦Cohort access and peer review
✦Priority tutor response
✦Early access to new projects
Monthly
$79/month
Cancel anytime
Join waitlist
30-day money-back guarantee
Already bought a project? Its price comes off the track.Teams and labs — hello@klerisy.com
Objections

The things you were
about to think.

You will write Python, but you are not asked to memorise syntax. The tutor explains a concept the moment it becomes relevant and never before. What you are being trained on is the decision, not the keystrokes.
Early access

Stop planning the pivot. Start the first project.

Waitlist members get early-access pricing, first pick of projects, and the first invitations when we open.

Loading application...