The words of data science & AI, in plain English.
266 terms, each in a sentence or two, with where our program teaches it, where to try it, and the interview questions that ask about it. The maths behind them is in the formula decoder.
266 terms
A
- A/A test
An A/B test of two identical versions, to check the set-up: at α = 0.05, about 5% will look significant by chance.
Analytics and experimentsTry it in the A/B test calculatorInterview questions
- A/B test
A randomised experiment comparing two versions on a metric, such as the conversion rate.
Analytics and experimentsTry it in the A/B test calculatorThe formulaInterview questions
- Absolute reference
A cell reference fixed with $ signs, like $A$1, so it doesn’t shift when the formula is copied.
Excel · Module 3Interview questions
- Accuracy
The share of predictions that are right.
Machine learning · Module 5The formulaTry it in the confusion matrixInterview questions
- Activation function
The non-linear function a neuron applies, such as ReLU or tanh; without one, stacked layers collapse into one straight-line map.
Deep learning · Module 6Try it in the network trainerInterview questions
- Adam
A popular optimiser that adapts each weight’s step using running averages of its gradients.
Deep learning · Module 6Try it in the network trainerInterview questions
- Agent memory
What an agent keeps between steps or sessions, such as a summary or stored facts.
AI agents · Module 10Interview questions
- Aggregate function
A function that turns many rows into one value: COUNT, SUM, AVG, MIN or MAX.
SQL · Module 4Interview questions
- AI agent
A system in which a model decides which steps to take and which tools to use, in a loop, to reach a goal.
AI agents · Module 10Interview questions
- ANOVA
Analysis of variance: tests whether the means of three or more groups differ.
Statistics · Module 2Interview questions
- API
An interface programs use to talk to each other, such as an address that returns predictions.
MLOps and production · Module 9Interview questions
- Approximate nearest neighbour (ANN)
Indexes, such as HNSW, that find close vectors fast, trading a little accuracy for speed.
RAG and search · Modules 8 and 9Interview questions
- Attention
How each token weighs, and mixes in, information from the other tokens.
LLMs and GenAI · Modules 7 and 8Lesson: AttentionThe formulaInterview questions
- AUC
The area under the ROC curve: the chance the model ranks a random positive above a random negative.
Machine learning · Module 5Try it in the confusion matrixInterview questions
B
- Backpropagation
Working out how the loss changes with every weight, by the chain rule, from the output back.
Deep learning · Module 6Try it in the network trainerInterview questions
- Batch and online inference
Predicting for many records on a schedule, or for each request as it arrives.
MLOps and production · Module 9Interview questions
- Batch normalisation
Normalising a layer’s inputs across each batch, so training is faster and steadier.
Deep learning · Module 6Interview questions
- Batch size
How many examples each step of training uses.
Deep learning · Module 6Interview questions
- Bayes’ theorem
Updates a probability with new evidence: P(A | B) = P(B | A) × P(A) ÷ P(B).
Statistics · Module 2Interview questions
- Bias-variance trade-off
Simple models miss patterns (bias); complex ones chase noise (variance); the best model balances the two.
Machine learning · Module 5Try it in the overfitting playgroundInterview questions
- Binomial distribution
The number of successes in a fixed number of yes-or-no trials, each with the same chance.
Statistics · Module 2Interview questions
- BM25
A keyword ranking formula that scores documents by the query’s words they contain, weighting rare words more.
RAG and search · Modules 8 and 9Try it in RAG chunkingInterview questions
- Byte-pair encoding (BPE)
Builds a tokenizer’s vocabulary by merging, again and again, the pair of symbols that appears most often in training text.
LLMs and GenAI · Modules 7 and 8Lesson: TokensInterview questions
C
- CALCULATE
The DAX function that works out an expression in a changed filter context.
Power BI · Module 3Interview questions
- Calculated column
A DAX column worked out row by row and stored in the model.
Power BI · Module 3Interview questions
- CASE WHEN
SQL’s if-then-else, inside a query.
SQL · Module 4Interview questions
- Causal mask
Stops each token attending to later ones, so a model writing text uses only what came before.
LLMs and GenAI · Modules 7 and 8Lesson: AttentionInterview questions
- Causation
When changing one thing changes another. Experiments, or careful causal methods, show it; a correlation alone doesn’t.
Statistics · Module 2Interview questions
- Central limit theorem
Averages of enough independent values tend to a normal distribution, whatever the shape of the values themselves.
Statistics · Module 2Try it in the central limit theorem simulatorInterview questions
- Chain of thought
Having a model work through its reasoning step by step before it answers.
LLMs and GenAI · Modules 7 and 8Interview questions
- Chi-square test
Tests whether counts in categories differ from what you’d expect, such as whether two categories are related.
Statistics · Module 2Interview questions
- Chunk overlap
Repeating text at the edges of chunks, so a sentence cut at one edge is whole in the next chunk.
RAG and search · Modules 8 and 9Try it in RAG chunkingInterview questions
- Chunking
Cutting documents into pieces for retrieval; the size and overlap decide what a search can find.
RAG and search · Modules 8 and 9Try it in RAG chunkingInterview questions
- Churn
Customers who stop buying or using a product; the churn rate is their share over a period.
Analytics and experimentsTry it in the confusion matrixInterview questions
- CI/CD
Continuous integration and delivery: testing and releasing changes automatically.
MLOps and production · Module 9Interview questions
- Citation
Pointing to the source passage behind each claim in an answer.
RAG and search · Modules 8 and 9Interview questions
- Class imbalance
When one class is much rarer than another, such as fraud; accuracy then misleads.
Machine learning · Module 5Try it in the confusion matrixInterview questions
- COALESCE
Returns the first value in its list that isn’t NULL, often used to fill gaps.
SQL · Module 4Interview questions
- Cohort analysis
Following groups that started at the same time, such as a sign-up month, to compare how they behave over time.
Analytics and experimentsInterview questions
- Concept drift
When the relationship between the inputs and the target changes, so a model’s learned rules go stale.
MLOps and production · Module 9Interview questions
- Conditional formatting
Formats cells by rules, such as colouring values above a target.
Excel · Module 3Interview questions
- Conditional probability
The chance of A given that B happened, written P(A | B).
Statistics · Module 2Interview questions
- Confidence interval
A range from a sample, made by a method that catches the true value a stated share of the time, such as 95%.
Statistics · Module 2The formulaTry it in the central limit theorem simulatorInterview questions
- Confusion matrix
A table of actual against predicted classes: true and false positives, and true and false negatives.
Machine learning · Module 5Try it in the confusion matrixInterview questions
- Context window
The most tokens a model can take in at once: the instructions, the conversation, any documents and its reply.
LLMs and GenAI · Modules 7 and 8Lesson: Context and the KV cacheInterview questions
- Conversion rate
The share of visitors who complete a goal, such as signing up or buying.
Analytics and experimentsInterview questions
- Convolutional neural network (CNN)
A network for images that slides small learned filters across its input.
Deep learning · Module 6Interview questions
- Correlation
How closely two variables move together, from −1 to 1 for a straight-line (Pearson) relationship. It doesn’t show cause.
Statistics · Module 2Interview questions
- Cosine similarity
How alike two embeddings are, by the angle between them: 1 is the same direction.
RAG and search · Modules 8 and 9The formulaLesson: EmbeddingsInterview questions
- Covariance
Whether two variables move together, in their combined units; correlation is covariance scaled to −1 to 1.
Statistics · Module 2Interview questions
- CROSS JOIN
Pairs every row of one table with every row of the other.
SQL · Module 4Interview questions
- Cross-validation
Splitting the data into k folds and training k times, each time testing on a different fold, for a steadier estimate.
Machine learning · Module 5Try it in the overfitting playgroundInterview questions
- CTE
A common table expression: a named subquery written with WITH, to make a query readable or to recurse.
SQL · Module 4Interview questions
D
- Dashboard
A single view of key metrics, built to be checked often.
Power BI · Module 3Interview questions
- Data drift
When the data a model sees in production changes from what it was trained on.
MLOps and production · Module 9Interview questions
- Data leakage
When information that won’t exist at prediction time slips into training, so results look better than they’ll be.
Machine learning · Module 5Interview questions
- Data model
The tables in a report and the relationships between them.
Power BI · Module 3Interview questions
- DataFrame
A pandas table of rows and labelled columns.
Python and pandas · Module 1Interview questions
- DAX
Data Analysis Expressions: the formula language of Power BI’s measures and calculated columns.
Power BI · Module 3Interview questions
- Decision threshold
The score above which a model predicts the positive class; moving it trades precision against recall.
Machine learning · Module 5Try it in the confusion matrixInterview questions
- Decision tree
A model that splits data by feature thresholds into branches that end in predictions.
Machine learning · Module 5Interview questions
- Decoder
A transformer that writes one token at a time, attending only to earlier tokens, such as GPT.
LLMs and GenAI · Modules 7 and 8Lesson: AttentionInterview questions
- Dictionary
Python’s mapping of keys to values: {"city": "Pune"}.
Python and pandas · Module 1Interview questions
- Dimension table
A table of attributes used to slice the facts, such as product, customer or date.
Power BI · Module 3Interview questions
- Docker
A tool that packages an app and everything it needs into a container that runs the same anywhere.
MLOps and production · Module 9Interview questions
- Dropout
Switching off random units during training, to reduce overfitting.
Deep learning · Module 6Interview questions
E
- Embedding
A learned list of numbers that stands for an item, such as a word or a product, so similar items get similar numbers.
Deep learning · Module 6Lesson: EmbeddingsInterview questions
- Embedding model
A model that turns text into embeddings for search, such as bge-small-en-v1.5.
RAG and search · Modules 8 and 9Lesson: EmbeddingsInterview questions
- Encoder
A transformer that reads a whole input at once, in both directions, such as BERT; suited to understanding and embedding text.
LLMs and GenAI · Modules 7 and 8Interview questions
- Ensemble
Combining several models’ predictions into a better one.
Machine learning · Module 5Interview questions
- Epoch
One full pass over the training data.
Deep learning · Module 6Interview questions
- Evaluation set
A fixed set of inputs with known good outputs, used to measure a model or prompt before and after a change.
MLOps and production · Module 9Interview questions
- Exception
An error raised while code runs, handled with try and except.
Python and pandas · Module 1Interview questions
- Experiment tracking
Recording each training run’s code, data, settings and results, with tools such as MLflow.
MLOps and production · Module 9Interview questions
- Exponential distribution
The waiting time between independent events that happen at a steady rate.
Statistics · Module 2Try it in the central limit theorem simulatorInterview questions
F
- F1 score
The harmonic mean of precision and recall, one number that balances them.
Machine learning · Module 5The formulaTry it in the confusion matrixInterview questions
- Fact table
The table of measurable events, such as each sale and its amount.
Power BI · Module 3Interview questions
- Feature
An input a model uses, such as tenure or the monthly charge.
Machine learning · Module 5Interview questions
- Feature engineering
Creating or reshaping features to help a model, such as a ratio or the day of the week.
Machine learning · Module 5Interview questions
- Feature importance
How much each feature contributes to a model’s predictions.
Machine learning · Module 5Interview questions
- Feature scaling
Putting features on similar scales, which distance-based models and models trained by gradient descent need.
Machine learning · Module 5Interview questions
- Few-shot prompting
Putting a few worked examples in the prompt to show the model the task.
LLMs and GenAI · Modules 7 and 8Interview questions
- Filter context
The filters in effect when a measure is worked out: slicers, rows, columns and the measure’s own.
Power BI · Module 3Interview questions
- Fine-tuning
Training a pretrained model further on your own data, for your task.
Deep learning · Module 6Interview questions
- Foreign key
A column that refers to another table’s primary key, linking the two tables.
SQL · Module 4Interview questions
- FULL OUTER JOIN
Keeps every row of both tables, matched where possible.
SQL · Module 4Interview questions
- Function
A named block of code that takes inputs, does a job and returns a result, so it can be reused.
Python and pandas · Module 1Interview questions
- Funnel
The steps users take towards a goal, and how many drop off at each.
Analytics and experimentsInterview questions
G
- Generator
A function that hands out values one at a time with yield, so it doesn’t hold them all in memory.
Python and pandas · Module 1Interview questions
- Gradient boosting
Trees built one after another, each correcting the errors of the ones before; XGBoost and LightGBM are popular versions.
Machine learning · Module 5Interview questions
- Gradient descent
Training by nudging every weight, step after step, in the direction that lowers the loss.
Deep learning · Module 6Try it in the network trainerInterview questions
- Greedy decoding
Always taking the likeliest next token.
LLMs and GenAI · Modules 7 and 8Lesson: Picking the next tokenInterview questions
- Grid search
Trying every combination of chosen hyperparameter values, scored by cross-validation.
Machine learning · Module 5Interview questions
- Grounding
Tying an answer to given sources, so it can be checked.
RAG and search · Modules 8 and 9Lesson: HallucinationInterview questions
- GROUP BY
Groups rows that share values, so aggregates like COUNT or SUM are worked out for each group.
SQL · Module 4Interview questions
- groupby
Splits a DataFrame into groups, applies a function to each and combines the results.
Python and pandas · Module 1Practise it in the Python LabInterview questions
- Grouped-query attention (GQA)
Query heads share key-value heads, which shrinks the KV cache with little loss of quality.
LLMs and GenAI · Modules 7 and 8Lesson: Context and the KV cacheInterview questions
- Guardrails
Checks on a model’s input and output that block unsafe, off-topic or wrongly formatted content.
AI agents · Module 10Interview questions
- Guesstimate
A structured estimate built from reasonable assumptions; interviews use them to test how you reason.
Analytics and experimentsInterview questions
H
- Hallucination
A fluent, confident answer that is false or unsupported.
LLMs and GenAI · Modules 7 and 8Lesson: HallucinationInterview questions
- HAVING
Filters groups after GROUP BY; WHERE filters rows before it.
SQL · Module 4Interview questions
- Human in the loop
A person reviewing or approving an AI system’s actions at key points.
AI agents · Module 10Interview questions
- Hybrid search
Combining keyword and semantic search results.
RAG and search · Modules 8 and 9Try it in RAG chunkingInterview questions
- Hyperparameter
A setting chosen before training, such as a tree’s depth or the learning rate.
Machine learning · Module 5Interview questions
- Hypothesis test
A way to judge whether data would be surprising if a default assumption, the null hypothesis, were true.
Statistics · Module 2Interview questions
I
- Index
A structure that speeds up lookups on a column, at the cost of storage and slower writes.
SQL · Module 4Interview questions
- INDEX and MATCH
A lookup pair: MATCH finds a position and INDEX returns the value there, in any direction.
Excel · Module 3Interview questions
- Inference
Running a trained model to get its outputs.
LLMs and GenAI · Modules 7 and 8Interview questions
- INNER JOIN
Keeps only the rows that have a match in both tables.
SQL · Module 4Interview questions
- Instruction tuning
Fine-tuning a model on instructions and good responses, so it follows requests.
LLMs and GenAI · Modules 7 and 8Interview questions
- Interquartile range (IQR)
The spread of the middle half of the data: the 75th percentile minus the 25th. Often used to flag outliers.
Statistics · Module 2Interview questions
J
- JOIN
Combines rows from two tables where a condition matches, such as the same customer ID.
SQL · Module 4Practise it in the SQL LabInterview questions
- Jupyter notebook
A document that mixes code, its output and notes, run cell by cell.
Python and pandas · Module 1Interview questions
K
- k-means
Clustering that assigns each point to the nearest of k centres, then moves the centres, until they settle.
Machine learning · Module 5Interview questions
- k-nearest neighbours
Predicts from the k training examples most similar to the new one.
Machine learning · Module 5Interview questions
- KPI
A key performance indicator: a metric a team tracks against a goal.
Analytics and experimentsInterview questions
- KV cache
The stored keys and values of earlier tokens, so each new token needn’t recompute them; it grows with every token.
LLMs and GenAI · Modules 7 and 8Lesson: Context and the KV cacheThe formulaInterview questions
L
- Label
The answer a supervised model learns to predict.
Machine learning · Module 5Interview questions
- LAG and LEAD
Read a value from the previous or next row in a window, such as to compare a month with the one before.
SQL · Module 4Interview questions
- Lambda
A small unnamed function written as one expression: lambda x: x + 1.
Python and pandas · Module 1Interview questions
- Large language model (LLM)
A neural network trained on vast amounts of text to predict the next token, used to write, answer and reason in language.
LLMs and GenAI · Modules 7 and 8Lesson: Picking the next tokenInterview questions
- Lasso
Linear regression with an L1 penalty, which can set coefficients to exactly zero and so selects features.
Machine learning · Module 5Interview questions
- Latency
How long a response takes; for LLMs, the time to the first token and the tokens per second.
MLOps and production · Module 9Lesson: Context and the KV cacheInterview questions
- Law of large numbers
As a sample grows, its average gets closer to the true average.
Statistics · Module 2Interview questions
- Learning rate
How big each step of gradient descent is: too big overshoots, too small crawls.
Deep learning · Module 6Try it in the network trainerInterview questions
- LEFT JOIN
Keeps every row of the left table, with the right table’s matches, or NULLs where there are none.
SQL · Module 4Interview questions
- Linear regression
Predicts a number as a weighted sum of the features plus a constant.
Machine learning · Module 5The formulaInterview questions
- List
Python’s ordered collection that can change: [1, 2, 3].
Python and pandas · Module 1Interview questions
- List comprehension
A compact way to build a list: [x * 2 for x in nums if x > 0].
Python and pandas · Module 1Interview questions
- LLMOps
MLOps for language models: prompts, evaluation, cost, speed and safety in production.
MLOps and production · Module 9Interview questions
- Log loss
The cross-entropy of predicted probabilities: a confident wrong answer costs the most.
Machine learning · Module 5The formulaInterview questions
- Logistic regression
Predicts the probability of a class by passing a weighted sum of the features through the sigmoid function.
Machine learning · Module 5Try it in the confusion matrixInterview questions
- Logits
A model’s raw scores for every token in its vocabulary, before softmax turns them into probabilities.
LLMs and GenAI · Modules 7 and 8Lesson: Picking the next tokenInterview questions
- LoRA
Low-rank adaptation: fine-tuning by training small added matrices instead of every weight.
LLMs and GenAI · Modules 7 and 8Interview questions
- Loss function
The number training tries to make small: how wrong the predictions are.
Deep learning · Module 6Interview questions
M
- Machine learning
Building models that learn patterns from data instead of following hand-written rules.
Machine learning · Module 5Practise it in the ML LabInterview questions
- Matplotlib
Python’s core library for charts.
Python and pandas · Module 1Interview questions
- MCP
The Model Context Protocol: an open standard for connecting AI applications to tools and data sources.
AI agents · Module 10Interview questions
- Mean
The average: add the values and divide by how many there are. A few extreme values pull it a long way.
Statistics · Module 2The formulaInterview questions
- Measure
A DAX calculation worked out in the current filter context, such as total sales for whatever is selected.
Power BI · Module 3Interview questions
- Median
The middle value once the values are sorted: half are above it, half below. A few extreme values barely move it.
Statistics · Module 2Interview questions
- merge
Joins two DataFrames on matching keys, like a SQL JOIN.
Python and pandas · Module 1Interview questions
- Minimum detectable effect
The smallest change a test is sized to detect reliably; halving it needs about four times the visitors.
Analytics and experimentsTry it in the A/B test calculatorInterview questions
- Missing values
Empty entries, NaN in pandas. Drop them, fill them or flag them, depending on why they’re missing.
Python and pandas · Module 1Interview questions
- MLOps
The practices for putting models into production and keeping them working: versions, deployment, monitoring and retraining.
MLOps and production · Module 9Interview questions
- Mode
The value that appears most often.
Statistics · Module 2Interview questions
- Model deployment
Making a trained model available to apps, such as behind an API.
MLOps and production · Module 9Interview questions
- Model monitoring
Watching a live model’s inputs, outputs and accuracy over time.
MLOps and production · Module 9Interview questions
- Model registry
A store of model versions and their status, from testing to production.
MLOps and production · Module 9Interview questions
- Multi-agent system
Several agents with different roles working together.
AI agents · Module 10Interview questions
- Multi-head attention
Several attention heads in one layer, each free to follow a different relationship.
LLMs and GenAI · Modules 7 and 8Lesson: AttentionInterview questions
- Multiple testing
Running many tests or metrics at once, which raises the chance of a false positive somewhere unless you adjust for it.
Analytics and experimentsInterview questions
N
- Neural network
Layers of simple units that weight their inputs, add them up and apply an activation, trained by gradient descent.
Deep learning · Module 6Try it in the network trainerInterview questions
- Neuron
One unit of a network: a weighted sum of its inputs plus a bias, passed through an activation.
Deep learning · Module 6Try it in the network trainerInterview questions
- Normal distribution
The symmetric bell curve: about 68% of values lie within one standard deviation of the mean, and about 95% within two.
Statistics · Module 2Try it in the central limit theorem simulatorInterview questions
- Normalisation
Organising tables so each fact is stored once, reducing duplicated data.
SQL · Module 4Interview questions
- North star metric
The one metric a company treats as the best sign of the value it delivers.
Analytics and experimentsInterview questions
- NULL
A missing or unknown value. Comparisons with NULL are unknown too, so test for it with IS NULL.
SQL · Module 4Interview questions
- Null hypothesis
The default assumption a test starts from, usually “no effect” or “no difference”.
Statistics · Module 2Interview questions
- NumPy
The Python library for fast arrays and the maths on them.
Python and pandas · Module 1Interview questions
O
- One-hot encoding
Turning a category into separate 0-or-1 columns, one for each category.
Machine learning · Module 5Interview questions
- Outlier
A value far from the rest, often more than 1.5 IQRs beyond the quartiles. It can be an error or a real, rare case.
Statistics · Module 2Interview questions
- Overfitting
When a model learns the noise in its training data and does worse on new data.
Machine learning · Module 5Try it in the overfitting playgroundInterview questions
P
- p-value
The chance of a result at least as extreme as yours if the null hypothesis were true. It is not the chance the null is true.
Statistics · Module 2The formulaTry it in the A/B test calculatorInterview questions
- pandas
The Python library for tables: reading, cleaning, joining and summarising data.
Python and pandas · Module 1Practise it in the Python LabInterview questions
- Parameters
A model’s learned numbers, its weights; Qwen 3 (8B) has 8.2 billion.
LLMs and GenAI · Modules 7 and 8Lesson: Context and the KV cacheInterview questions
- PARTITION BY
Splits rows into groups for a window function, like GROUP BY without collapsing the rows.
SQL · Module 4Interview questions
- PCA
Principal component analysis: finds the directions in which data varies most, to describe many features with a few.
Machine learning · Module 5Lesson: EmbeddingsInterview questions
- Peeking
Checking a running test again and again and stopping at the first significant result, which inflates false positives.
Analytics and experimentsTry it in the A/B test calculatorInterview questions
- Percentile
The value below which a given share of the data falls: the 90th percentile is higher than 90% of values.
Statistics · Module 2Interview questions
- Perplexity
How surprised a language model is by a text; lower means it predicts the text better.
LLMs and GenAI · Modules 7 and 8Interview questions
- Pivot table
Summarises a table by grouping rows and columns, with sums, counts or averages.
Excel · Module 3Interview questions
- Planning
An agent breaking a goal into steps before it acts.
AI agents · Module 10Interview questions
- Poisson distribution
The number of events in a fixed interval when they happen independently at a steady average rate, like calls a minute.
Statistics · Module 2Interview questions
- Population
The whole group you want to know about.
Statistics · Module 2Interview questions
- Positional encoding
Tells a transformer where each token sits, since attention on its own ignores order.
LLMs and GenAI · Modules 7 and 8Interview questions
- Power Query
The tool in Excel and Power BI for getting and cleaning data in recorded, repeatable steps.
Power BI · Module 3Interview questions
- Precision
Of the cases predicted positive, the share that really are.
Machine learning · Module 5The formulaTry it in the confusion matrixInterview questions
- Precision-recall curve
Precision against recall at every threshold; more telling than ROC when positives are rare.
Machine learning · Module 5Try it in the confusion matrixInterview questions
- Prefill and decode
The two phases of writing: reading the prompt in one pass, then adding one token a pass.
LLMs and GenAI · Modules 7 and 8Lesson: Context and the KV cacheInterview questions
- Primary key
A column, or set of columns, that identifies each row of a table uniquely.
SQL · Module 4Interview questions
- Prompt
The text given to a model: instructions, examples and the input.
LLMs and GenAI · Modules 7 and 8Interview questions
- Prompt injection
Instructions hidden in the input, such as a web page or a document, that try to take over a model.
AI agents · Module 10Interview questions
- Python
A general-purpose programming language, and the one most used in data science.
Python and pandas · Module 1Practise it in the Python LabInterview questions
- PyTorch
A popular Python library for building and training neural networks.
Deep learning · Module 6Interview questions
Q
- Quantisation
Storing a model’s numbers in fewer bits, such as 8 or 4, to cut memory and speed it up for a small loss in quality.
LLMs and GenAI · Modules 7 and 8Interview questions
- Query plan
How the database will run a query; EXPLAIN shows it, to find the slow steps.
SQL · Module 4Interview questions
- Query, key and value
The three lists each token makes for attention: what it looks for, what it offers to be found by, and what it passes on.
LLMs and GenAI · Modules 7 and 8Lesson: AttentionInterview questions
R
- R²
The share of the target’s variance a regression model explains.
Machine learning · Module 5The formulaInterview questions
- RAG
Retrieval-augmented generation: finding the relevant passages first, then giving them to the model to answer from.
RAG and search · Modules 8 and 9Try it in RAG chunkingLesson: HallucinationInterview questions
- Random forest
Many decision trees, each on a random sample of rows and features, averaged; usually steadier and more accurate than one tree.
Machine learning · Module 5Interview questions
- RANK and DENSE_RANK
Rank rows, giving ties the same rank; RANK then skips numbers, DENSE_RANK doesn’t.
SQL · Module 4Interview questions
- Re-ranking
Scoring the top retrieved passages again with a stronger model before using them.
RAG and search · Modules 8 and 9Interview questions
- ReAct
A pattern in which a model alternates reasoning and actions, using each result to decide the next step.
AI agents · Module 10Interview questions
- Recall
Of the real positives, the share the model finds; also called sensitivity.
Machine learning · Module 5The formulaTry it in the confusion matrixInterview questions
- Recurrent neural network (RNN)
A network that reads a sequence step by step, carrying a hidden state; LSTMs are one kind.
Deep learning · Module 6Interview questions
- Regularisation
Penalising complexity to reduce overfitting, such as L1 (lasso) or L2 (ridge).
Machine learning · Module 5Try it in the overfitting playgroundInterview questions
- Reinforcement learning
Learning by trial and error, from rewards and penalties.
Machine learning · Module 5Interview questions
- Relational database
Data stored in tables of rows and columns, linked by keys.
SQL · Module 4Interview questions
- ReLU
An activation that keeps positive values and turns negative ones to zero: max(0, x).
Deep learning · Module 6Try it in the network trainerInterview questions
- Retention
The share of users who come back or stay active after a given time.
Analytics and experimentsInterview questions
- Ridge regression
Linear regression with an L2 penalty, which shrinks the coefficients.
Machine learning · Module 5Try it in the overfitting playgroundInterview questions
- RLHF
Reinforcement learning from human feedback: tuning a model towards the answers people prefer.
LLMs and GenAI · Modules 7 and 8Interview questions
- RMSE
Root mean squared error: a typical size of a regression model’s errors, in the target’s units.
Machine learning · Module 5Try it in the overfitting playgroundInterview questions
- ROC curve
Recall against the false-positive rate at every threshold.
Machine learning · Module 5Try it in the confusion matrixInterview questions
- ROW_NUMBER
Numbers the rows 1, 2, 3 within each partition, with no ties.
SQL · Module 4Interview questions
- Row-level security
Rules that limit which rows each person can see in a report.
Power BI · Module 3Interview questions
S
- Sample
The part of a population you actually measure.
Statistics · Module 2Interview questions
- Sample size
How many observations a study needs. For an A/B test it depends on the rate now, the effect to detect, α and power.
Analytics and experimentsTry it in the A/B test calculatorInterview questions
- Sampling bias
When a sample doesn’t represent the population, so its results don’t carry over; surveying only your app’s users about the app is one.
Statistics · Module 2Interview questions
- Seasonality
A pattern that repeats at regular intervals, such as festival-season sales.
Analytics and experimentsInterview questions
- Self join
Joining a table to itself, such as matching employees with their managers in the same table.
SQL · Module 4Interview questions
- Self-attention
Attention within one sequence: each token looking at the others, or at the earlier ones.
LLMs and GenAI · Modules 7 and 8Lesson: AttentionInterview questions
- Semantic search
Searching by meaning, by comparing embeddings rather than words.
RAG and search · Modules 8 and 9Try it in RAG chunkingLesson: EmbeddingsInterview questions
- Series
One labelled column of values in pandas.
Python and pandas · Module 1Interview questions
- SHAP values
Explain a prediction by each feature’s share of it, based on Shapley values from game theory.
Machine learning · Module 5Interview questions
- Sigmoid
Squeezes any number into the range 0 to 1, used for the probability of one of two classes.
Deep learning · Module 6Interview questions
- Significance level (α)
The p-value below which you reject the null, often 0.05: the false-positive rate you accept.
Statistics · Module 2Interview questions
- Simpson’s paradox
A trend that shows within groups reverses or disappears when the groups are combined.
Statistics · Module 2Interview questions
- Skewness
How lopsided a distribution is. Right-skewed data has a long tail of large values, so its mean usually sits above its median.
Statistics · Module 2Interview questions
- Slicer
A filter on a report page that viewers click.
Power BI · Module 3Interview questions
- SMOTE
Creating synthetic examples of the rare class to balance the training data.
Machine learning · Module 5Interview questions
- Softmax
Turns a list of scores into probabilities that add up to 1.
Deep learning · Module 6The formulaLesson: Picking the next tokenInterview questions
- SQL
Structured Query Language: the language for querying and changing data in relational databases.
SQL · Module 4Practise it in the SQL LabInterview questions
- Standard deviation
The square root of the variance, in the data’s own units: a typical distance of a value from the mean.
Statistics · Module 2The formulaInterview questions
- Standard error
The standard deviation of an estimate, such as a sample’s mean: for a mean, the spread divided by √n.
Statistics · Module 2The formulaTry it in the central limit theorem simulatorInterview questions
- Star schema
A model with one table of events, such as sales, linked to tables that describe them, such as products and dates.
Power BI · Module 3Interview questions
- Statistical power
The chance a test detects an effect of a given size if it’s real; 80% is a common target.
Statistics · Module 2Try it in the A/B test calculatorInterview questions
- Structured output
Getting a model to answer in a fixed format, such as JSON that matches a schema.
LLMs and GenAI · Modules 7 and 8Interview questions
- Subquery
A query inside another query.
SQL · Module 4Interview questions
- SUMIFS and COUNTIFS
Add up, or count, the values that meet several conditions.
Excel · Module 3Interview questions
- Supervised learning
Learning from labelled examples to predict a label, such as whether a customer will leave.
Machine learning · Module 5Interview questions
- Support vector machine
A classifier that finds the boundary with the widest margin between the classes.
Machine learning · Module 5Interview questions
- System prompt
Instructions set by the app, before the user’s messages, that shape how a model behaves.
LLMs and GenAI · Modules 7 and 8Interview questions
T
- t-test
Compares means (one sample, two samples or paired) when the spread is estimated from the data.
Statistics · Module 2Interview questions
- Temperature
Divides the scores before softmax: below 1 sharpens the choice of next token, above 1 flattens it.
LLMs and GenAI · Modules 7 and 8The formulaTry it in the temperature playgroundInterview questions
- Test set
Data held back to measure how a model does on data it hasn’t seen.
Machine learning · Module 5Interview questions
- Token
A piece of text, often a word or part of one, that a model reads and writes as a number.
LLMs and GenAI · Modules 7 and 8Lesson: TokensTry it in the token counterInterview questions
- Tokenizer
Cuts text into tokens and turns them into numbers, and back again.
LLMs and GenAI · Modules 7 and 8Lesson: TokensTry it in the token counterInterview questions
- Tool calling
A model asking the app to run a function, such as a search, a calculator or an API, and using the result.
AI agents · Module 10Interview questions
- Top-k sampling
Drawing the next token only from the k likeliest.
LLMs and GenAI · Modules 7 and 8Lesson: Picking the next tokenTry it in the temperature playgroundInterview questions
- Top-p sampling
Drawing only from the smallest set of tokens whose chances add up to p; also called nucleus sampling.
LLMs and GenAI · Modules 7 and 8Lesson: Picking the next tokenTry it in the temperature playgroundInterview questions
- Training set
The data a model learns from.
Machine learning · Module 5Interview questions
- Transaction
A group of changes that succeed or fail together; ACID: atomic, consistent, isolated and durable.
SQL · Module 4Interview questions
- Transfer learning
Starting from a model trained on one task and adapting it to another.
Deep learning · Module 6Interview questions
- Transformer
The neural network design behind LLMs, built on attention.
LLMs and GenAI · Modules 7 and 8Lesson: AttentionInterview questions
- Tuple
An ordered collection that can’t change once it’s made.
Python and pandas · Module 1Interview questions
- Type I error
A false positive: rejecting a null hypothesis that is true.
Statistics · Module 2Interview questions
- Type II error
A false negative: missing an effect that is real.
Statistics · Module 2Interview questions
U
- Underfitting
When a model is too simple to catch the pattern, and does badly on training and new data alike.
Machine learning · Module 5Try it in the overfitting playgroundInterview questions
- UNION
Stacks the results of two queries: UNION removes duplicate rows, UNION ALL keeps them.
SQL · Module 4Interview questions
- Unsupervised learning
Finding structure in data without labels, such as groups of similar customers.
Machine learning · Module 5Interview questions
V
- Validation set
Data used to choose between models and settings, kept apart from the test set.
Machine learning · Module 5Interview questions
- Vanishing gradient
When gradients shrink as they pass back through many layers, so the early layers barely learn.
Deep learning · Module 6Interview questions
- Variance
How spread out values are: the average squared distance from their mean.
Statistics · Module 2The formulaInterview questions
- Vector database
A database that stores embeddings and quickly finds the ones nearest a query.
RAG and search · Modules 8 and 9Interview questions
- Vectorisation
Applying an operation to a whole array at once instead of looping, which is far faster in NumPy and pandas.
Python and pandas · Module 1Interview questions
- View
A saved query that can be used like a table.
SQL · Module 4Interview questions
- Virtual environment
A separate set of Python packages for one project, so projects don’t clash.
Python and pandas · Module 1Interview questions
- vLLM
An open-source engine for serving LLMs fast, which manages the KV cache’s memory in pages.
MLOps and production · Module 9Interview questions
- VLOOKUP
Finds a value in the first column of a range and returns a value from a column to its right.
Excel · Module 3Interview questions
- Vocabulary
The fixed list of tokens a model knows; GPT-4o’s tokenizer has about 200,000.
LLMs and GenAI · Modules 7 and 8Lesson: TokensInterview questions
W
- WHERE
Filters rows before any grouping.
SQL · Module 4Interview questions
- Window function
Works across related rows without collapsing them, such as a running total or a rank, written with OVER (…).
SQL · Module 4Practise it in the SQL LabInterview questions
X
- XLOOKUP
Excel’s newer lookup: searches any column and returns from any other, with an exact match by default.
Excel · Module 3Interview questions
Z
- z-score
How many standard deviations a value sits from the mean.
Statistics · Module 2The formulaInterview questions
No term matches. Try fewer letters, or .
Short answers, then the real thing.
A sentence or two
Each term is said the way you’d explain it to a friend, accurate enough to repeat in an interview.
Try it, don’t just read it
Where a term has a tool, a lab or a lesson, it links there: drag a threshold, train a network, count your own tokens.
Then the interview
Every term links to the interview questions on its topic, with answers to say out loud.