Ritesh Modi — ExplorablesInteractive explainers you learn by messing with.https://www.loopingly.com/2026-08-22T00:00:00ZRitesh Modihttps://www.loopingly.com/aboutPrompting techniques, and when to use eachhttps://www.loopingly.com/explorables/prompting-techniques2026-08-22T00:00:00Z2026-08-22T00:00:00ZEleven prompting techniques against one task. Fence off pasted text so it cannot give orders, force a shape, sample five times and count the answers.Predicting the next characterhttps://www.loopingly.com/explorables/predicting-the-next-character2026-08-22T00:00:00Z2026-08-22T00:00:00ZFill a 625-box count grid by hand, smooth it so unseen pairs stop breaking it, then train an eight-number embedding and watch the loss come down.Every control on the Playgroundhttps://www.loopingly.com/explorables/playground-controls2026-08-22T00:00:00Z2026-08-22T00:00:00ZEvery control on the OpenAI Playground, one at a time. Move reasoning effort, verbosity and text format, and watch what each one changes in the request.How twenty companies run generative AIhttps://www.loopingly.com/explorables/how-twenty-companies-run-generative-ai2026-08-22T00:00:00Z2026-08-22T00:00:00ZTwenty production deployments across six layers. Resize the retrieval chunks until they ruin an answer, turn the queue dial, and find each system's scar.A history of generative AI, 1948 to nowhttps://www.loopingly.com/explorables/history-of-generative-ai2026-08-22T00:00:00Z2026-08-22T00:00:00ZPlay Shannon's 1948 letter-guessing game, drag n across a bigram table until counting collapses, then watch backpropagation pick up where it failed.A machine that produces instead of choosinghttps://www.loopingly.com/explorables/generative-ai2026-08-22T00:00:00Z2026-08-22T00:00:00ZOld classifiers picked an answer off a list. Watch a sentence become numbers, then a spread of odds, and make the sampling choice for yourself.Medallion architecture on ten rows a dayhttps://www.loopingly.com/explorables/medallion-ten-rows-a-day2026-08-18T00:00:00Z2026-08-18T00:00:00ZTen rows a night through bronze, silver and gold. Append the raw file, merge one row per order, rebuild the report, then drop a late row in and watch it move.The Delta table APIhttps://www.loopingly.com/explorables/delta-table-api-explained2026-08-18T00:00:00Z2026-08-18T00:00:00ZA table here is a folder of files plus a log saying which ones count. Overwrite, merge, time travel and vacuum, watching the folder change at every call.Classification, regression, deep learninghttps://www.loopingly.com/explorables/three-motions-of-a-model2026-08-08T00:00:00Z2026-08-08T00:00:00ZTwenty-two logged commutes, three models, one loop. Nudge a rule downhill, hold rows back, drag a threshold, then bend a straight line with two hidden layers.Four numbers disagree about one modelhttps://www.loopingly.com/explorables/regression-metrics-by-hand2026-08-08T00:00:00Z2026-08-08T00:00:00ZAdd ten gaps one at a time, then drop a cracked-screen laptop into the test set and watch MAE, MSE, RMSE and R squared disagree about the very same model.Fifteen numbers, learnedhttps://www.loopingly.com/explorables/neural-network-by-hand2026-08-08T00:00:00Z2026-08-08T00:00:00ZFifteen weights, all of them on the page. Walk one row forward through two hidden layers, watch the error walk back, then turn the learning rate up until it explodes.One slider, three different modelshttps://www.loopingly.com/explorables/classification-metrics-by-hand2026-08-08T00:00:00Z2026-08-08T00:00:00ZDrag one threshold across fifteen emails and watch precision, recall and the four confusion counts swing while accuracy sits at 0.87 and tells you nothing.From a word to a vectorhttps://www.loopingly.com/explorables/03-words-to-vectors2026-08-07T00:00:00Z2026-08-07T00:00:00ZSpin a compass between two words and watch the angle turn into a similarity score, then point it at ten real vectors and pull out each word's nearest neighbours.What is intelligence?https://www.loopingly.com/explorables/what-is-intelligence2026-08-01T00:00:00Z2026-08-01T00:00:00ZEvery definition of intelligence lets in something that is not, and shuts out something that is. Break four of them yourself, then answer the harder questions.A language model made of matchboxeshttps://www.loopingly.com/explorables/matchbox-language-model2026-08-01T00:00:00Z2026-08-01T00:00:00ZIn 1960 a stack of matchboxes learned noughts and crosses using coloured beads. Point the same machine at words and train a language model by hand, bead by bead.How data becomes an equationhttps://www.loopingly.com/explorables/how-data-becomes-an-equation2026-08-01T00:00:00Z2026-08-01T00:00:00ZEight days of a lemonade stand, worked by hand into a line, a cluster and a policy — supervised, unsupervised and reinforcement learning from one table.The three types of machine learninghttps://www.loopingly.com/explorables/three-types-of-machine-learning2026-07-31T00:00:00Z2026-07-31T00:00:00ZThree ways to learn to cook, in the same order as the three kinds of machine learning. What separates them is one question: when do you find out you were wrong?How the AI words nesthttps://www.loopingly.com/explorables/the-layers-of-intelligence2026-07-31T00:00:00Z2026-07-31T00:00:00ZAI, machine learning, neural nets and deep learning nest like India, Maharashtra, Mumbai, Bandra. Then the diagram keeps going, and GenAI stops fitting inside.When to use a rule instead of a modelhttps://www.loopingly.com/explorables/never-use-ai-when-rules-will-do2026-07-31T00:00:00Z2026-07-31T00:00:00ZWhen you can make the promise, do not place the bet. Slide the error tolerance until the model stops being the cheaper answer and a plain rule wins outright.How models learnhttps://www.loopingly.com/explorables/how-models-learn2026-07-31T00:00:00Z2026-07-31T00:00:00ZA model on a hillside in fog can only feel which way the ground tilts. Set the learning rate yourself and watch gradient descent converge, crawl, or diverge.AI vs machine learning vs automationhttps://www.loopingly.com/explorables/ai-vs-ml-explained2026-07-31T00:00:00Z2026-07-31T00:00:00ZAutomation, machine learning and AI all watch, decide and act. Flip one thermostat between all three and see exactly where rules stop and learning starts.What AI is good and bad athttps://www.loopingly.com/explorables/what-ai-is-good-and-bad-at2026-07-26T00:00:00Z2026-07-26T00:00:00ZThere is one clean line between what AI does well and badly, and it is not the one most people guess. Learn to call the green and red lights before you hit send.Tokenizationhttps://www.loopingly.com/explorables/tokenization2026-07-26T00:00:00Z2026-07-26T00:00:00ZA tokenizer is a frozen record of what was frequent in one pile of text. Type a sentence, watch it get cut and priced, then break it with an unusual word.The regulatory atlashttps://www.loopingly.com/explorables/regulatory-atlas2026-07-26T00:00:00Z2026-07-26T00:00:00ZGDPR to the EU AI Act: ten frameworks, who each binds, what each demands, and the controls that satisfy them. Filter by obligation to see where they overlap.How reasoning models workhttps://www.loopingly.com/explorables/reasoning-models-explainer2026-07-26T00:00:00Z2026-07-26T00:00:00ZAsk a reasoning model something hard and it pauses. Open up that pause: what it does in the gap, how it learned to, and when the extra tokens are wasted money.API rate limitinghttps://www.loopingly.com/explorables/rate-limiting-explainer2026-07-26T00:00:00Z2026-07-26T00:00:00ZRate limiting is refusing work cheaply so you can keep doing work at all. Starve, feed and flood a live token bucket, and catch the moment it starts returning 429.Prompt engineeringhttps://www.loopingly.com/explorables/prompt-engineering-explainer2026-07-26T00:00:00Z2026-07-26T00:00:00ZClimb from a plain question to ReAct agents on one stubborn problem, and watch exactly which change of wording moves the answer and which changes nothing.Postgres table partitioninghttps://www.loopingly.com/explorables/postgres-partitioning-explainer2026-07-26T00:00:00Z2026-07-26T00:00:00ZDrag a project id and watch the planner skip seven of eight partitions, then see why retention stops being a slow DELETE and becomes an instant DROP.Partition equal subset sumhttps://www.loopingly.com/explorables/partition-equal-subset-sum2026-07-26T00:00:00Z2026-07-26T00:00:00ZCan a pile of numbers split into two stacks of equal weight? Tip the scale with your finger, then watch the same question solved as a subset-sum table.Multimodal AIhttps://www.loopingly.com/explorables/multimodal-explainer2026-07-26T00:00:00Z2026-07-26T00:00:00ZA phone pointed at a foreign menu needs sight and language at once. Build the shared space that lets one model mix its senses, and see how multi-model differs.What an AI assistant actually doeshttps://www.loopingly.com/explorables/is-it-actually-thinking2026-07-26T00:00:00Z2026-07-26T00:00:00ZShort answer: no — it is autocomplete that read the internet. See the one trick underneath, and two of the biggest beginner worries about AI fall away at once.The GRC registers an organization keepshttps://www.loopingly.com/explorables/grc-operating-registers2026-07-26T00:00:00Z2026-07-26T00:00:00ZThe registers an organisation keeps on itself — risk, control, asset, incident, exception, vendor. What belongs in every row, and who has to sign it.Governing generative AIhttps://www.loopingly.com/explorables/grc-for-genai2026-07-26T00:00:00Z2026-07-26T00:00:00ZRun governance for generative AI end to end through one organisation's real control library, risk register and compliance posture — not a checklist.Governance, risk and compliance explainedhttps://www.loopingly.com/explorables/grc-explained2026-07-26T00:00:00Z2026-07-26T00:00:00ZFrameworks, policies, controls, risk, mitigation, compliance — six words that sound like bureaucracy but describe one system. Trace a single rule end to end.The seven building blocks of GRC: a lessonhttps://www.loopingly.com/explorables/grc-building-blocks-lesson2026-07-26T00:00:00Z2026-07-26T00:00:00ZThe same seven GRC blocks, taught rather than listed — worked through one organisation in order, showing how a single failing block breaks the six others.The seven building blocks of GRChttps://www.loopingly.com/explorables/grc-building-blocks2026-07-26T00:00:00Z2026-07-26T00:00:00ZA reference for the seven pieces every GRC programme is built from: what each block is, what it is not, and the question it answers when an auditor asks.Generative AI vocabularyhttps://www.loopingly.com/explorables/genai-vocabulary-explainer2026-07-26T00:00:00Z2026-07-26T00:00:00ZToken, embedding, context window, RAG, agent — the GenAI vocabulary in the order the ideas depend on each other. Click any term to watch it work.Evaluating and comparing AI modelshttps://www.loopingly.com/explorables/evaluating-models-explainer2026-07-26T00:00:00Z2026-07-26T00:00:00ZTwo chatbots, two answers, no way to choose by gut. Build a small eval set, score both models against it, and watch a coin flip turn into a measurement.Finding and tracking AI modelshttps://www.loopingly.com/explorables/discovering-models-trends-explainer2026-07-26T00:00:00Z2026-07-26T00:00:00ZA new model lands every week and the announcements all sound alike. Read a model card the way engineers do, then track which releases actually mattered.Azure service principals and OAuthhttps://www.loopingly.com/explorables/azure-service-principals-explainer2026-07-26T00:00:00Z2026-07-26T00:00:00ZBuild an Azure service principal from scratch, decode the JWT it is handed, then widen its role assignment and watch the blast radius grow along with it.Leading AI transformationhttps://www.loopingly.com/explorables/ai-transformation-explainer2026-07-26T00:00:00Z2026-07-26T00:00:00ZEfficiency projects pay back once; reinvention changes what the business sells. Move the sliders to see which one your AI portfolio is really funding.