Coding is one thing. But can you explain it?

The Inferential is an adaptive learning system for data science and machine learning concepts. Built for practitioners who want to stay sharp or prepare for interviews.

How it works
α
2.0
β
3.0
Trials
0
Mastery
40
TheGradient.ai represents your mastery of a concept as a beta distribution — not a point estimate. As you review more concepts and answer more practice questions, the system uses Bayesian statistics to update its belief about your mastery of the concept, causing your distribution to shift and tighten intuitively.
Adaptive learning

Review key concepts.

An intelligent concept scheduler balances breadth and depth. It focuses on what you need to improve on, but also makes sure you stay fresh on what you already know.

Intelligent scheduling

The scheduler dynamically prioritizes concepts that need work and concepts that are up for review. But it also tracks which specific dimensions of each concept you need to work on — intuition, assumptions, trade-offs, failure modes, edge cases, misconceptions. The front of each card prompts you with a concept and a specific dimension.

  • The scheduler prioritizes concepts you don't know well and concepts that you haven't seen for a while.
  • It also prioritizes areas where it doesn't have much evidence yet about your mastery level.
  • The scheduler learns continuously — the more concepts you review, the more personalized it becomes.

Adaptive learning

The back of each card shows a concise overview of the concept and dimension you're being tested on — plus your current mastery distribution. You rate how well you know that dimension, and the scheduler updates its belief about how well you know that concept, subcategory, and category. Your mastery distributions update immediately, and the scheduler uses that information to decide what to show you next.

  • Rate each dimension/concept: "Got it", "Learning it", or "Missed it".
  • Your mastery distributions update in real time after every response.
  • The mean of the distribution shifts to reflect the scheduler's new belief about your mastery.
  • The credible intervals for the distribution narrow as the scheduler accumulates evidence of your mastery.

All dimensions in context

Below each card, you can always read concise overviews of all the other dimensions of the concept you've just been tested on — so you can keep the whole picture in mind while still focusing on one specific angle.

  • See concise overviews of all dimensions without interrupting your review.
  • Looking at the other dimensions doesn't affect their mastery distributions.
  • Quick access to the other dimensions helps you orient yourself to new concepts — if you don't know a concept very well yet, you can remind yourself of the definition or intuition.

Focus your session with filters

You can focus your review session at any time by choosing specific categories or subcategories to review. The scheduler works the same way — focusing on the areas where you're weak while still reviewing things you know well — just within the bounds you set. It's really useful when you need to prepare for a particular interview topic or want to consolidate a weak area before moving on.

  • Filter by category or subcategory to focus your review session.
  • Within the areas you've choosen to focus on, the scheduler still prioritizes what needs the most work.
  • Switch focus areas freely without losing your progress.
Broad coverage

300+ concepts. Multiple dimensions.

The Inferential covers a wide variety of key concepts across data science and machine learning — curated, organized, and ready to review. No deck-building, no imports.

300+ curated concepts

Supervised and unsupervised learning, deep learning, probabilistic methods, optimization, evaluation, and more — organized by category and subcategory. No deck-building required.

  • Supervised & unsupervised learning
  • Deep learning & neural architectures
  • Probabilistic methods & Bayesian inference
  • Optimization, evaluation, and deployment

Concept dimensions

Definition, intuition, assumptions, trade-offs, edge cases, and misconceptions — all individually tracked. The scheduler knows exactly which angle needs work.

  • Definition & intuition
  • Assumptions & trade-offs
  • Edge cases & misconceptions

Breadth and depth

The scheduler downweights concepts you've already seen in your active session, so you don't review the same card twice in one sitting.

  • Tracks your current session automatically
  • No repeated back-to-back reviews
  • Maximizes coverage per session

Streak tracking

Track daily practice consistency to keep memory decay at bay and build the habit that makes long-term retention possible.

  • Daily activity tracking
  • Visual streak history
  • Keeps decay from compounding

Mastery decay

Memory fades when it isn't reinforced. Your estimated mastery gradually decreases on unreviewed concepts, returning them to your queue when they drift below your threshold.

  • Concepts you've locked in decay slowly
  • Shaky topics decay faster
  • Decay rate is concept-specific, not a flat interval

Flexible filters

Shift your focus areas at any time without losing historical mastery data on topics you've already studied.

  • Change focus areas freely
  • Historical data always preserved
  • Mastery restored when you return
Mastery tracking

Track what you've mastered.

See intuitive mastery distributions at every level — dimension, concept, subcategory, and category. Track your activity and look up concepts you want to review.

Activity tracking

Your mastery page shows your activity over time and your mastery distributions for every category, subcategory, concept, and dimension. It tracks how many cards you review each day and your daily streak.

  • It keeps tracks of how many concepts you review each day and how many you've seen at least once.
  • It estimates your overall mastery across all concepts.
  • Tracking your daily streak helps you to remember to review regularly.

Dimension-level mastery

For each concept, the learning system tracks your mastery of relevant dimnesions. The number of dimensions varies by concept — not every dimension is relevant to every concept. You can hover over any concept to see exactly which dimensions you're strong on — and which need work.

  • Every dimension of every concept has its own mastery distribution.
  • The dimensions vary by concept — it depends on what's relevent.
  • If you haven't reviewed a dimension yet, the concept's mastery card clearly shows that your distribution for that dimension is based on a prior — the learning system's initial belief about how well you know it based on your self-reported familiarity with each subcategory.
  • You also get a mastery distribution for every concept, subcategory, and category so you can easily see what you need to focus on.

Onboarding calibration

At sign-up, you rate your familiarity with each subcategory. This sets your starting mastery distributions so the scheduler doesn't waste your time on material you already know well.

  • You rate your familiarity with each subcategory before you start using the app, and the learning system uses that information to form initial beliefs about your mastery levels.
  • As you review cards, the learning system will update these initial beliefs based on your responses.
  • You can update your familiarity with any subcategory at any time — your earned mastery is always preserved.
  • If you down-rate a subcategory, but have already demonstrated some mastery of a concept in that subcategory, you don't lose that progress.

Look up any concept

Your mastery page doubles as a reference guide. Select any concept to read concise overview for all of its dimensions — useful when you want to look up something specific without starting a new review session.

  • You can read all content for any dimension of any concept at any time — without starting a review session.
  • See the content for each dimension right next to its mastery distribution.
  • Seach for any category, subcategory, or concept to instantly surface their mastery distributions.
Interview preparation

From recognition to fluency.

Reviewing concepts builds recall. Practice questions build the ability to reason and communicate under pressure — what interviews actually test. Available in Pro+.

Conceptual questions

Test whether you can explain, compare, or reason about concepts — not just recognize it. Conceptual questions ask you to articulate ideas in your own words, the way a technical interview actually would.

  • There's one conceptual question for every dimension of every concept.
  • Answering questions requires active reasoning, not passive recall.
  • The back of each card shows a concise model answer to the question so you can self-assess your response.
  • Answering conceptual questions affects your mastery distributions — just like reviewing concepts does.

Scenario questions

Apply your knowledge to realistic situations — diagnose a failing pipeline, choose between architectures, explain a result to a stakeholder. Scenario questions help you articulate and defend a response to a practical scenario.

  • There's one scenario question for every dimension of every concept.
  • Scenario questions ask you to apply your knowledge of key concepts — to respond to (sometimes dubious) claims by colleagues, to diagnose common production feailures, to make choices that require weighing trade-offs.
  • Answering scenario questions affects your mastery distributions — just like reviewing concepts does.
Recommended posts
Adaptive Learning

Review. Learn. Practice.

Whether you want to stay fresh, expand your knowledge, or prepare for interviews, The Inferential can help you improve your mastery of key data science and machine learning concepts.

Get Started