Identity 01
The Hebrew Root (Eshcol)
Eshcol is Hebrew for a cluster of grapes. Many grapes, one stem. That is where our name comes from.
Eskolx Labs
We rebuild statistical and machine learning libraries from scratch in Python so you understand them line by line.
Scroll to beginExplore the reposAI writes code you don't understand. We give you the hard problems that force understanding. You rebuild the statistical libraries everyone takes for granted, then turn them on questions nobody has answered.
The name
Why the Name Eshcol
Identity 02
The ‘X’ Factor
The X stands for three things.
- eXecution: theory becomes running code.
- eXploration: methods and approaches probed deeper and deeper until we reach the frontier.
- Scale: tools that start as a single script and grow into full research pipelines.
Identity 03
From Paper to Library
We read the paper first, then write the code, then test it. Each function starts from a book or paper, gets a first working version, gets checked against the standard libraries, and only then joins our library.
The Four Phases
Twelve weeks, four phases. Each phase builds on the last: statistical primitives, then modeling, then advanced methods, until the libraries can carry real research.
PHASE 1
Foundational Statistics & Probability
Weeks 1-12. Descriptive statistics, elementary probability distributions, and hypothesis testing, written as pure Python with no black boxes. At the end of the three months, participants turn their own packages loose on novel research.
- Descriptive statistics
- Elementary probability distributions
- Hypothesis testing
PHASE 2
Modeling & Classical Machine Learning
After the basics, we move on to modeling and classical machine learning built on what we grew ourselves. Assumption checks will be wired end to end. Every method should be textbook-backed or paper-backed.
- Modeling on our own packages
- Assumption checks
- Feature diagnostics
- Reference comparison against the famous implementations
PHASE 3
Advanced Applied Machine Learning
Next comes advanced data science, still from scratch. You read the papers, then write tested, documented code you can explain line by line.
- Neural networks
- Modern architectures
- Time series & stationarity
PHASE 4
Toward Automated Data Science
The long game. Libraries and an open knowledge base good enough that data science starts automating itself, built by people who rebuilt the basics first.
- Analysis pipelines
- Spatial statistics
- Knowledge-base-driven methods
Phase one is running now.
Twelve weeks, everyone on the same phase at the same time. Read the plan, and write to us if you want in.
From seed to harvest
Community Channels
Follow the code, the research, and the conversation across every channel.
GitHub
github.com/eskolx-labs
Open-source repositories & codebase docs
Open GitHub
Telegram
t.me/eskolx_labs
Community chat & builder updates
Open Telegram
Discord
discord.gg/BKhMFznJa
Builder chat & live help
Open Discord
linkedin.com/company/eskolx-labs
Technical announcements & talent recruitment
Open LinkedIn
Join us
Requirements & FAQ
How to join, how we work, where the project stands. No sales copy, just the answers.
What we do
- Build statistical and data-science libraries from scratch, in pure Python
- Read the books and papers behind every method before writing a line
- Publish everything under MIT, including the study notes
- Use our own packages on real, unsolved problems
What we don't
- Serve as your first Python tutorialArrive knowing the basics. We start at real problems, not at print statements.
- Promise financial rewardMaybe someday, not today. What you leave with is skill, shipped work, and your name on public commits.
- Ban AIBuild with it if you like. You will explain every line and understand everything you shipped.
- Be a training platformNo lectures, no grades, no certificates. You build real packages with a small team, and the work is the point.
- Hand you production-ready libraries todayWe build from scratch to learn. The packages are MIT-licensed and public, but early-stage.
What we ask of you
Four requirements, and that is the whole filter.
01
What you bring
You bring working Python and a real interest in understanding statistics and machine learning at a deeper level. You should already know the main statistical and machine learning packages in Python. Some object-oriented programming helps. You should also want to read long technical books and research papers, because every function starts there.
02
Notes go public
You explain what you build in Obsidian notes written from books and papers, plus explanatory diagrams drawn in tldraw. Your notes land in our open knowledge base under MIT with your name on them as author. Start with the open study vault.
03
Small, fast teams
We keep a small group of steady contributors who work with us in short cycles. We can be demanding. That said, anyone from anywhere in the world can contribute to what we build.
04
Finish by teaching
The loop ends when you can explain what you built. The notes and tldraw diagrams you make are part of that. The best ones get published here on our site.
01What you bring
02Notes go public
03Small, fast teams
04Finish by teaching
Scroll — the bar reads top to bottom
Where the project stands
- Status
- Actively maintained. Last push Sep 6, 2026.
- Current goal
- Three months, basic statistical packages: descriptive statistics, elementary probability distributions, and hypothesis testing.
- Releases
- Cut when a package milestone closes. The statistical basics land first.
- License
- MIT. Everything public.
Asked often
Anyone interested in statistics, data science, or machine learning who would rather build than watch. Most of our people are students. We give you hard technical work that forces a deeper understanding of how methods work.
Using a library and understanding one are different skills. Ours start naive, get compared against the famous implementations, and improve until they are solid. That comparison is the work. We also do it for the pure fun of seeing how things work underneath.
We do not reinvent random number generation or the fast array and tensor operations in NumPy and TensorFlow. We use those packages for basic operations. The statistical and machine learning methods themselves we build ourselves, line by line.
Introduction to Probability and Statistics for Engineers and Scientists, sixth edition, by Sheldon M. Ross. We work through it in order and turn its methods into tested Python code.
Serious but open. You need working Python, familiarity with the main Python packages for statistics and machine learning, and some object-oriented programming helps. You do not need advanced class statistics. Curiosity about books and papers matters more. The hard parts arrive as projects, not prerequisites.
Yes. The code is MIT licensed, the study vault downloads for free, and nobody pays tuition.
Books and research papers. Participants take notes before implementing anything, and those notes live in an open Obsidian vault anyone can download. Explanations good enough for the public end up on this site.
Not yet. This one stays short on purpose. Until the expanded version ships, ask us directly on Telegram or by email.
Something unanswered? Telegram or eskolxlabs@gmail.com.