Learn AI by doing

Practical, video-first AI courses for students, employees, freelancers, business owners, and companies ready to use AI at work.

New to AI? Start here.

Pick one practical course and learn at your own pace. These beginner courses are available in the Academy catalogue now.

• Claude Cowork & AI Knowledge Work - Use the selected demonstrations to study knowledge-work patterns: organizing files, adding durable context, researching a website, packaging reusable capabilities, drafting business artifacts, and comparing flows with agents. The emphasis is on inspectable inputs and reviewable outputs, because interfaces and permissions can change.
• AI Image Creation: Prompts, Consistency, and Local Tools - Work from a visual brief to a reviewable image workflow. You will compare generation and editing approaches, practice prompt structure, preserve character and scene continuity, and test hosted and local options without treating a particular interface as permanent. The course emphasizes version records, permissions, artifact review, and disclosure for synthetic media.
• AI Content and Marketing Workflows - Build a small content system rather than a pile of tool tips. You will practice prompt and context design, multi-output briefs, bounded automation, UGC-style drafts, mobile shortcuts, marketing experiments, and controlled prompt labs. Each workflow keeps verification, privacy, claims review, and human approval visible so that a changing product UI does not become the course’s only source of truth.
• Read the AI Handbook alongside your course

Learn AI by doing

Practical, video-first AI courses for students, employees, freelancers, business owners, and companies ready to use AI at work.

• SQL Analytics with Window Functions - Develop practical fluency with SQL window functions by moving from the OVER clause and partitioning to ranking, comparative calculations, and a more extended lesson on window-function analysis. Learn to distinguish a window calculation from row-collapsing aggregation, choose an ordering that makes ties and frame boundaries explicit, and validate results against small examples. The original assignments use invented or permitted data and emphasize deterministic ordering, clear grain, and careful interpretation; a query result is not meaningful until its population and assumptions are stated.
• Model Drift Monitoring: Detect, Diagnose, and Improve - Connect statistical drift signals with production-model diagnosis and governed improvement. Distinguish input-distribution change from change in predictive relationships, combine data-quality and outcome evidence, and plan a response that does not equate an alert with a reason to retrain. Fictional assignments emphasize baselines, time-aware evaluation, subgroup review, rollback, and ownership.
• Grounded Answers: Build and Evaluate a RAG Workflow - Move from document retrieval to a reviewable question-answering workflow. Learners compare local PDF-based systems, examine a self-checking retrieval approach, and plan an evaluation that tests retrieval and generation separately. Original activities emphasize source traceability, privacy, failure analysis, and human review rather than treating a fluent answer as proof.
• Generative AI: A First Practical Guide - A short starting path for people who have never studied generative AI. Four independently licensed demonstrations introduce the idea, explain the limits of generated output, and show writing and reading uses. Build a small, checked workflow rather than treating fluent output as a reliable source.
• Advanced Image Segmentation: Semantic, Instance, and Panoptic Methods - A focused computer-vision progression from segmentation framing to semantic labels, instance and panoptic representations, and point/pixel segmentation with tracking. Learners distinguish prediction units, annotation conventions, evaluation choices, and temporal association, then design task-specific assessments for boundaries, class imbalance, and identity consistency. Selected uploads have English audio; study notes and assignments are original analysis, not transcripts.
• Parameter-Efficient LLM Fine-Tuning: PEFT, LoRA, and QLoRA - Examine parameter-efficient fine-tuning from its motivation and low-rank adaptation through PEFT workflows and quantized LoRA. Compare frozen base weights with trainable adapters, make target modules and data choices explicit, and design evaluations separating task gains from memorization, quantization effects, and deployment costs. Original analysis accompanies English-audio uploads; it is not transcript text.
• Continual Learning: Memory, Stability, and Plasticity - Examine continual-learning scenarios, stability–plasticity trade-offs, memory and generalization, and a task-agnostic continual reinforcement-learning study. The sequence moves from defining sequential-task evaluation to adaptation mechanisms and memory trade-offs, then considers a simple baseline in a reinforcement-learning setting. Learners distinguish retention from transfer and design evaluations that expose forgetting, task-order effects, and resource costs. Videos have English audio; these independently written notes and assignments do not reproduce transcripts.
• Reinforcement Learning: Value Methods and Policy Gradients - A progression from Markov decision processes and reinforcement-learning objectives through value and policy families, fitted value iteration, and deterministic policy gradients. Learners connect the choice of method to action spaces, data collection, and stability concerns, while distinguishing algorithmic ideas from guarantees of successful control. The selected uploads have English audio; these original study notes and assignments do not reproduce video transcripts.
• Applied Machine Learning Projects: Prepare, Predict, and Compare - Use a sequence of applied examples to connect data preparation, a supervised classification project, clustering, and time-series forecasting. The course focuses on decisions that make a result interpretable: define the prediction target, inspect and prepare data without leaking future information, compare a result with a suitable baseline, and describe what the model cannot establish. The original activities use safe, public or synthetic data and emphasize reproducibility, uncertainty, and responsible communication rather than treating a tutorial result as general evidence.
• A/B Testing: Design, Run, and Interpret Experiments - Develop a practical experiment workflow that connects a product question to a clearly defined treatment, comparison group, and outcome. Learners consider random assignment, exposure and measurement, practical significance, and the risks of reading too much into a noisy result. The original activities use fictional or appropriately authorized aggregate data; they emphasize predeclared decisions, guardrails, and honest uncertainty rather than promising that any test will produce a definitive answer.
• Information Retrieval: Click Models, Relevance, and Learning to Rank - This advanced sequence moves from linguistic relevance modeling and click-based interaction signals to online pairwise learning and deep ranking architectures used in web-scale systems. Learners distinguish relevance judgments from observed clicks, examine feedback and exposure bias, and connect ranking objectives to operational evaluation. The focus is search scoring and ranking methodology—not retrieval-augmented generation. The English-audio recordings are accompanied by original analytical notes and assignments, not copied transcripts.
• Transformer Architectures: Attention to Time Series - A technical progression from transformer fundamentals and a transparent spreadsheet-scale implementation to attention variants and the Informer time-series architecture. Learners connect architectural choices to sequence structure, computational costs, and forecasting assumptions, then propose an evaluation that avoids temporal leakage. The selected videos are English-language uploads; original study guidance and assignments emphasize analysis rather than reproducing any video's transcript.
• Prompt Engineering: Build Clearer Requests - A hands-on introduction to communicating with generative AI. Four independently rights-checked videos provide a progression from prompt basics and prompt quality to examples of prompting techniques and clearer written requests. Practise with a low-stakes task, compare outputs, and treat every answer as a draft that needs review.
• Practical Data Analysis with Python and Pandas - Build a practical analysis workflow in short steps: prepare a notebook, understand Python values, load common files, select and join data, summarize groups, and inspect a dataset. The selected lessons are open-license demonstrations from Alex The Analyst; they are not a substitute for current library documentation. Learners finish with a reproducible exploratory-analysis notebook and a clear record of assumptions.
• Graph Neural Networks: Message Passing and Graph Learning - Explore graph-attention foundations, neural message passing, node and graph classification, and multi-relational knowledge-graph completion. The sequence connects graph structure and feature aggregation to prediction tasks, then asks how augmentation and geometric assumptions affect generalization. Learners make graph construction choices explicit and design leakage-aware evaluation; the English-audio videos are paired with independently written study materials, not transcripts.
• Evaluating Generative Models: Metrics, Benchmarks, and Trust - Move from output-quality criteria to benchmark limitations, trust evidence, and a multimodal evaluation case. Rather than treating a leaderboard or single score as a verdict, learners define intended use, inspect data and rubric fit, examine failure slices, and decide what human review is still required. The lessons are English-language uploads; all accompanying learning notes and assessments are original.
• Self-Supervised Learning for Graph and Network Representations - Explore self-supervised graph representation learning through adaptive contrastive augmentation, multi-channel hypergraph modeling, contextual embeddings for heterogeneous link prediction, and logical-query representations in knowledge graphs. The course compares what different pretext signals make learnable and how graph type, augmentation, geometry, and downstream evaluation shape the result. Learners design leakage-aware tests rather than treating an unlabeled objective as evidence of general-purpose representations. English-audio videos are paired with independently written study materials, not transcripts.
• Advanced Object Detection: DETR and Data-Efficient Learning - Build a focused sequence on object detection: establish task and evaluation context, reason about DETR-style set prediction, examine transformer detector design, and consider probabilistic active learning for annotation selection. Distinguish per-image localization and classification from temporal tracking; emphasize appropriate metrics, data splits, annotation quality, and compute-aware comparisons. Original notes accompany English-audio videos without reproducing transcripts.
• Agentic Workflows: Plan, Use Tools, and Keep Human Control - Study the transition from a single response to a multi-step agent workflow. Learners map planning, tool calls, and handoffs; compare implementation approaches; and set boundaries for actions that could affect people, data, or external systems. The course's original assignments use fictional scenarios and make approval, logging, and failure recovery explicit.
• Python Data Workflows: Collect, Transform, and Analyze - Follow a practical data workflow from a public web page or API to structured data, numerical operations, and an analysis of relationships. Learners consider permission, respectful access, schema and type checks, missingness, and reproducible transformations before interpreting results. The course's original assignments distinguish correlation from causation and ask learners to document provenance and uncertainty. Examples are for learning and do not authorize scraping a site or collecting data that its owner has not made available for that use.
• Federated Learning: Aggregation, Personalization, and Representations - Examine the federated-learning setup, FedAvg aggregation, and methods that personalize models or learn federated representations. The course focuses on non-identical client distributions, uneven participation, communication, and the privacy limits of model exchange. Learners design evaluations distinguishing global performance from client utility and avoid equating local data storage with formal privacy. Original study notes are not transcripts.
• Knowledge Graph Completion: Representations, Embeddings, and Evaluation - This advanced course focuses on knowledge-graph completion rather than general graph neural networks. Learners connect structural and text features with entity–relation embeddings, examine efficient non-sampling objectives and schema-aware instance completion, and scrutinize evaluation protocols for link prediction. Across the sequence, attention stays on candidate construction, leakage, ranking metrics, and whether an evaluation reflects the intended use. The selected videos have English audio; original study materials are analytical guidance, not copied transcripts.
• Product Experimentation: Analysis, Trade-offs, and Platforms - Connect experiment interpretation to product decision-making and the systems that coordinate tests at scale. The course uses a paradox example, a lecture on models and A/B testing, a product discussion about when testing is appropriate, and a platform-management session to examine different kinds of evidence and operational trade-offs. Learners practice checking segmentation, overlapping experiments, governance, and measurement quality; no platform or presenter claim is treated as universal proof that an experiment is valid.
• MLOps Observability: Monitor Production Systems - Explore observability from metrics collection and alert design to production handoffs and notebook-to-service reliability. Separate infrastructure signals from evidence about predictions; examine privacy-aware instrumentation, bounded labels, ownership, reproducible releases, and recovery planning. Activities use fictional services and do not claim that a monitoring stack guarantees reliable outcomes.
• Your First Data Analysis Workflow - An orientation to three common ways to inspect tabular data, not a promise to master all three in one sitting. Four open-license demonstrations show an Excel project, MySQL setup, a SQL exploration, and cleaning with Pandas. Follow one path with a small dataset and document assumptions before moving to the others.
• Check Before You Trust: Fact-Checking AI Output - A beginner pathway for treating AI-generated claims as items to investigate rather than facts to repeat. Four rights-checked videos address hallucinations, a verification demonstration, human review, and practical ways of identifying faulty output. Learners practise separating claims, locating independent evidence, and clearly marking what remains uncertain.
• Geospatial Analysis: Spatial Data, Queries, and Joins - Build a careful workflow for querying and joining geographic features, then connect desktop GIS operations with a spatial database. Learners distinguish geometry predicates, coordinate reference systems, and the grain of joined records; they also examine how boundary choices and source provenance shape an analysis. The assignments use invented or appropriately licensed geographic examples, require checks for invalid or mismatched geometries, and avoid treating a map or spatial association as causal evidence. Verify source terms and coordinate metadata before reusing any real dataset.
• Machine Learning Foundations: Algorithms to Pipelines - Connect the geometric intuition of vectors to nearest-neighbor reasoning, boosted trees, loss functions, preprocessing pipelines, and two compact neural-network examples. The sequence uses open-license CampusX lessons as worked demonstrations. It emphasizes asking what data, objective, split, and evaluation measure a model uses, rather than treating a code example as evidence that a model will generalize.
• Quantitative Finance: Backtesting and Market Data - Examine how financial data, strategy rules, and simulation choices combine in a backtest. Learners distinguish a portfolio backtest from a single-strategy example, trace a Python-oriented implementation workflow, and scrutinize whether a test avoids look-ahead bias, survivorship bias, and unrealistic execution assumptions. The course treats attention-grabbing claims as claims to investigate, not established results. Its assignments use hypothetical data and stress transaction costs, time-aware validation, benchmark comparisons, risk measures, and the limits of extrapolating historical performance. This course is educational and is not investment advice.
• Generative Image Workflows: Configure, Create, and Review - Explore how desktop interfaces, model sources, and text-to-image workflows fit together, then consider what can and cannot be inferred from an image. Learners plan safe experiments, track model and prompt provenance, and review output for rights, disclosure, and misleading use. These original activities avoid claiming that a particular interface is current or that image detectors can conclusively identify provenance. Only use models, prompts, and images you are permitted to use.
• Bayesian Optimization for Efficient Experimentation - Move from the motivation for Bayesian optimization to surrogate-based experimental design and automated performance tuning. Learners reason about uncertainty, acquisition choices, noisy objectives, and practical constraints such as evaluation cost and parameter domains. The course emphasizes reproducible comparisons with random search and careful reporting of the search budget; original notes support the English-audio videos without copying transcripts.
• Differential Privacy: Sensitivity, Metrics, and Private Statistical Analysis - Build foundations in differential privacy, examine metrics related to local sensitivity, compare formal privacy definitions, and study private multi-party sketching for large-scale statistics. Emphasis is on neighboring-dataset assumptions, composition, utility, and the distinction between a mathematical guarantee and implementation evidence. Original study notes are not transcripts.
• Anomaly Detection: Streaming, Multivariate, and Interpretable Methods - This advanced course moves from anomaly detection in streaming data to multi-aspect streams, multivariate service indicators, and interpretable outlier analysis. Learners distinguish a detection score from a verified incident, examine temporal and cross-variable context, and assess how explanations should be validated with domain knowledge. The sequence includes an HTM-focused anomaly discussion alongside research presentations on streaming and multivariate methods; it does not assume that one detector suits every operational setting. The English-audio materials have original study notes and assignments, not copied transcripts.
• Responsible AI: Privacy, Ethics, and Review - An introductory course on using generative AI with care for people and information. Its four open-license selections move from practical privacy considerations to ethics, responsible-AI foundations, and research integrity. Original study activities help learners assess what may be shared, who could be affected, and when human review or a safer alternative is needed.
• Applied Recommender Systems: Books, Movies, and Playlists - Use distinct recommendation settings to reason about interaction-based and content-based prototypes, then assess what a compact demonstration can establish. Define feedback events and ranking goals, inspect item attributes and exposure bias, and plan consent, diversity, cold-start, and user-control checks. Assignments use invented cases or permitted catalog metadata; they do not claim that a demo predicts universal taste or proves product impact.
• Secure Machine Learning: Private Inference and Model Inversion - Explore privacy-preserving machine-learning systems through a progression from a general framework to secure nearest-neighbor classification and quantized neural-network evaluation, then examine model inversion as a concrete privacy threat in collaborative filtering. The course distinguishes computational protection from formal privacy guarantees and asks how threat models, leakage channels, and utility costs should be evaluated. The final attack-focused lesson is treated as security analysis, not as endorsement of extracting personal data. Videos have English audio; accompanying notes and assignments are original and do not reproduce transcripts.
• Interpretable Machine Learning: Explanations and Limits - Explore the distinction between interpretability and post-hoc explanations, examine feature attribution, and consider interpretable neural models and production constraints. Learners test explanation stability, check whether a method reflects model behavior, and communicate an explanation's scope to the people who need it. The course pairs English-language Creative Commons videos with original analytical notes and assignments.
• Data Pipeline Orchestration: Schedulers and DAGs - Explore data workflow scheduling through a Luigi pipeline talk, a comparison of Airflow and Luigi, a visual DAG-composer example, and migration lessons from enterprise schedulers. Learners connect task dependencies to retries, idempotence, observability, and change management rather than treating a scheduler as a substitute for reliable task code. The activities are tool-neutral and ask for explicit recovery and ownership plans; the selected talks are examples, not claims that one orchestration product is best for every team.
• Evaluate and Deploy AI Applications Responsibly - Connect application-level evaluation with the practical work of packaging and exposing a model-backed service. Learners define representative tests, examine reproducibility and API boundaries, and create release gates that include security, rollback, and human ownership. These independently written activities complement the videos; they are not a promise that any specific framework or deployment recipe remains current.
• Multi-Object Tracking: Identity Association and Robust Trackers - Study multi-object tracking as temporal association rather than per-frame detection. Progress from tracking concepts and association decisions to GHOST as a concrete tracker example. Analyze identity switches, missed and false detections, motion and appearance cues, and sequence-level evaluation without assuming a method generalizes beyond tested conditions. English-audio videos are supported by original notes, not copied transcripts.
• Deep Learning and Computer Vision with CNNs - Move from the role of deep learning and neural-network families to PyTorch fundamentals, convolutional structure, pretrained models, and transfer learning. The final lesson supplies a bounded image-classification project context. The material is presented through open-license CampusX demonstrations, so learners should verify framework APIs and dataset terms against current documentation before implementing a project.
• Repeatable Prompt Design: Structure, Test, and Review - Treat a prompt as one component of a larger application rather than a magic phrase. Learners define task, context, constraints, output format, and checks; compare prompt revisions against stable examples; and evaluate whether a proposed prompt-writing tool actually improves results. The course does not endorse claims in any video title or promise that a particular template will work across models. Its original assignments emphasize repeatability, evidence, privacy, and human judgment.
• Recommendation Algorithms: Collaborative Filtering in Practice - Move from the recommender ranking problem to user-based neighborhoods, weighted candidate scores, and item-based similarity. Trace assumptions in interaction matrices, consider overlap and exposure, and compare methods with popularity baselines using ranking quality, coverage, cold-start tests, and sensitivity checks. Assignments use tiny invented matrices and emphasize that offline scores do not prove user benefit, fairness, or causal product impact.
• Bayesian Inference: Learning, Posterior Computation, and MCMC - Progress from Bayesian inference and learning to posterior computation with Markov chain Monte Carlo. Learners articulate prior, likelihood, and posterior assumptions, distinguish inference targets from computational approximations, and assess how sampling diagnostics and model specification shape conclusions. Emphasis is placed on uncertainty and sensitivity analysis rather than treating a numerical posterior as automatically trustworthy. Original notes are not transcripts.
• Efficient Neural Inference: Quantization and Compression - Study neural-network quantization, a learned step-size quantization method, inference efficiency, and a broader set of large-language-model efficiency techniques. Learners distinguish representation and compute trade-offs from hardware outcomes, compare compression methods without assuming they are interchangeable, and design task-specific quality and latency checks. The selected videos have English audio; accompanying original material is analytical guidance rather than copied transcript text.
• Recommender Systems: Collaborative Filtering, Factorization, and Debiasing - Explore recommender systems from collaborative filtering and alternative preference signals to bit-level similarity coding, fairness-aware recommendation, and factorization machines. The sequence emphasizes how feedback is collected, what user-item interactions encode, and how ranking quality and fairness depend on evaluation design. Treat each method as a research proposal to test against transparent baselines, not a universal solution. The English-audio videos pair with original notes, not copied transcripts.
• Causal Inference: Experiments, Matching, and Regression - A progression from causal questions and assumptions to randomized experiments, matching, and regression adjustment. Learners define a target effect, draw a causal diagram, identify confounding and selection risks, and communicate the limits of observational conclusions. The material is advanced and methodological: examples support learning, not a substitute for domain expertise or a claim that any one estimator solves identification.
• NLP Systems: From Language Structure to Transformers - Build a practical conceptual bridge from text processing and sequence labels to neural language models and transformer representations. Learners compare task framing, data choices, and model outputs, then plan evaluation that respects context and error costs. Assignments use invented or public text and emphasize that a model's fluent output is not a substitute for checking task fit or representation bias.
• Data Engineering: ETL, Reverse ETL, and Analytics Pipelines - Follow four complementary views of pipeline work: a modern design overview, a Databricks ETL implementation, reverse ETL orchestrated with Airflow, and an Airflow/PySpark analytics pipeline. Learners distinguish ingestion and transformation choices from operational delivery into downstream systems, then document contracts, lineage, failure recovery, and privacy boundaries. The course treats architecture and vendor demonstrations as illustrative; it does not imply that a particular stack or movement of data is appropriate without source permissions and destination controls.
• AI Image Creation: From Prompt to Review - A beginner-oriented route through core ideas and practical approaches to generative images. Four independently verified Creative Commons videos introduce image-generation concepts, demonstrate image creation, and explore prompt formats and a beginner workflow. Learners create an original low-stakes brief, compare results, and review image quality, provenance, consent, and intended use before sharing.
• Few-Shot Learning: Generalization Across Knowledge, Events, and Graphs - Begin with the few-shot learning setup, then examine low-example methods for knowledge validation, taxonomy-aware event detection, and molecular graph property prediction. The course connects support examples and task structure to adaptation, while emphasizing that success on a small benchmark does not guarantee transfer to new domains or label distributions. Learners compare against simple baselines, account for label and task variation, and evaluate generalization at the appropriate entity or domain level. The videos are in English and the accompanying learning materials are original, not transcripts.
• NLP Foundations to Transformers and LLMs - Build an NLP mental model in sequence: define the task, place preprocessing and representation in a pipeline, try a supervised text-classification framing, then connect self-attention to the historical development of language models. The selected CampusX lessons are instructional source material rather than a promise of a particular model outcome. Exercises focus on making preprocessing choices and evaluation assumptions visible.
• Excel Data Analysis: From Cleanup to Charts - Learn a practical sequence for answering a small data question in Excel: inspect and clean a table, calculate useful fields, summarize categories with a pivot table, and present a chart that fits the question. The four lessons use distinct, rights-checked demonstrations. Work on a copy of a permitted dataset, retain the original, and explain what your final summary does and does not show.
• Build GenAI Apps with LangChain and LangGraph - Start with the role of LangChain in generative-AI applications, then examine components, model interfaces, prompts, structured outputs, and parsers before moving to a LangGraph agentic workflow. CampusX provides the selected demonstrations and code context; package names and APIs can change, so treat the exercises as patterns to verify rather than immutable production instructions. The course rewards explicit schemas, test inputs, and observable state transitions.
• Introductory SQL: Ask Questions of Tables - This beginner sequence builds an exploratory query in small, reviewable steps. Begin by selecting columns from a table, narrow the records with a row condition, summarize the filtered rows by category, and then distinguish filters on source rows from conditions on grouped results. The lessons are demonstrations, not a promise of a particular database setup; use a safe practice table and adapt dialect-specific details to your environment.
• AI Automation & Agent Operations - Study practical automation patterns without treating a demo as a production blueprint. The course moves from an end-to-end content workflow through remote operation, scheduling, extensions, supervisor-style multi-agent designs, and an n8n workflow with human quality control. Learners are expected to test permissions, failure recovery, data handling, and change-sensitive tool steps.
• Python and Pandas: A First Table Analysis - A focused introduction to working with tabular data in Python using Pandas. Start with the DataFrame as a labeled table, load an external file, filter rows or columns to match a question, and summarize groups. The selected videos come from their named uploading channels and are independently rights-checked. Practice with a non-sensitive file, verify each transformation, and record assumptions rather than treating code output as self-validating.
• Claude Cowork & AI Knowledge Work - Use the selected demonstrations to study knowledge-work patterns: organizing files, adding durable context, researching a website, packaging reusable capabilities, drafting business artifacts, and comparing flows with agents. The emphasis is on inspectable inputs and reviewable outputs, because interfaces and permissions can change.
• Power BI Fundamentals: Import, Model, Visualize - Follow a practical path through Power BI Desktop: get oriented with a first visualization, prepare an imported table, understand how table relationships affect analysis, and choose visuals that communicate a checked result. Each lesson is a separate Creative Commons-licensed demonstration from its uploader. Power BI features and interfaces can change, so treat screen details as version-specific and document your data and modeling decisions.
• Applied Prompting for Research & High-Stakes Work - Move from prompt structure and examples to legal-process context, conversation-gap analysis, retrieval grounding, visual research, and keyword discovery. The course treats model output as a draft for inspection—not as legal advice, a finding of fact, or a substitute for qualified review. Each assignment produces a prompt, evidence record, or evaluation artifact that can be audited.
• Machine Learning: Core Ideas and Data Splits - A four-lesson entry path from the basic idea of machine learning to learning-task categories, model concepts, and data splitting. Learners compare supervised, unsupervised, and reinforcement-learning examples, then examine why training and evaluation data must be separated carefully. The final activity applies these ideas to a small, permissioned dataset; no programming background is required.
• LLM Foundations: Deep Learning, Models & Retrieval - Connect neural-network training and transfer-learning practice with language-model internals, model evaluation, prompting, agent loops, MCP tool connections, and retrieval-oriented workflows. The material spans long-form technical lessons and shorter demonstrations; use the assignments to distinguish durable concepts from version-sensitive commands and to test every system boundary.
• Neural Networks: From Neurons to Learning - An accessible four-lesson introduction to neural networks, moving from the artificial neuron to layered networks, activation functions, and learned representations. The course uses conceptual explanations rather than requiring code or calculus. Learners finish by drawing a tiny network and explaining why its outputs still need testing on examples beyond those used to fit it.
• AI & LLM Foundations to Research Agents (Urdu-English) - Build a conceptual path from machine learning and NLP to large language models, pretrained model tooling, prompting, and a Python research-agent capstone. The sequence is presented in Urdu-English instruction and should be used as a foundation for experiments, not as a promise that a model API, library, or hosted service will remain unchanged. Assignments emphasize source handling, structured outputs, and human review.
• Data Visualization: Principles, Perception, and Encodings - A four-lesson beginner course about designing and reading data visualizations. It begins with the purpose of visualization, then considers perceptual organization, visual principles, and visual encodings such as position, length, and color. Learners practise matching a chart to a question and annotating uncertainty and limitations instead of treating a polished graphic as proof.
• AI Image Creation: Prompts, Consistency, and Local Tools - Work from a visual brief to a reviewable image workflow. You will compare generation and editing approaches, practice prompt structure, preserve character and scene continuity, and test hosted and local options without treating a particular interface as permanent. The course emphasizes version records, permissions, artifact review, and disclosure for synthetic media.
• Data Ethics: Privacy, Use, and Responsible Linkage - A practical introduction to ethical reasoning throughout a data lifecycle. Learners identify affected people and purposes, consider responsible use and privacy in learning analytics, then examine how combining datasets can create new risks even when individual sources seem harmless. The course provides a review framework—not legal advice—and ends with a concrete data-use decision record.
• AI Video Production: From Image-to-Video to Cinematic Ads - Plan short AI video work as shots, tests, and review checkpoints. The sequence moves from text-to-video prompting through cinematic planning, image animation, image-to-video experiments, local setup and troubleshooting, and an ad-oriented capstone. Durable production habits—briefs, continuity sheets, provenance records, defect logs, and human approval—remain useful as tools and models change.
• No-Code AI Automation: Build and Review a Workflow - A practical introduction to AI-assisted automation for people who do not program. The four licensed demonstrations move from an n8n beginner overview to a first AI automation, a Gmail-focused example, and a longer Make.com masterclass. Build along only with non-sensitive sample data: product interfaces and integrations change, and an automation should be tested and reviewed before it can affect real people or records.
• AI Content and Marketing Workflows - Build a small content system rather than a pile of tool tips. You will practice prompt and context design, multi-output briefs, bounded automation, UGC-style drafts, mobile shortcuts, marketing experiments, and controlled prompt labs. Each workflow keeps verification, privacy, claims review, and human approval visible so that a changing product UI does not become the course’s only source of truth.
• AI Writing Workflows: From Idea to Revision - Build a repeatable writing process without handing over editorial responsibility. These four Creative Commons demonstrations cover ChatGPT-supported brainstorming, a focused email prompt, a short lesson on making AI content feel more human to its audience, and reverse outlining with AI summaries. The course treats generated text as an editable draft: learners supply their own verified facts, check claims and tone, and make the final decisions.
• AI Research and Productivity Systems - Design research and productivity routines as auditable workflows. The lessons cover workspace permissions, meeting capture, scheduled research agents, product-signal collection, feature evaluation, agent security, and voice-assisted work. Rather than promising automatic accuracy, the course asks learners to define source boundaries, check outputs, protect sensitive information, and retain a manual fallback.
• AI Research Workflows: Find, Verify, Organize - A beginner path for using AI research features without mistaking generated summaries for evidence. The videos introduce AI deep research and research tools, examine verification of AI research, and show a research-reading workflow in Roam Research. Learners practise defining a question, locating original sources, recording source details, and marking what remains uncertain. AI tools can omit or misrepresent material; the researcher remains responsible for checking every important claim.
• Responsible AI: Fairness, Explainability, Privacy, and Trust - Learn to turn responsible-AI concerns into concrete review questions and decision records. The sequence moves from ethical scoping and competing fairness definitions to bias-test design, explanation limits, privacy threat modeling, copyright and provenance questions, and deepfake verification. It does not provide legal advice or a universal fairness score; it emphasizes evidence, affected people, uncertainty, human oversight, and escalation.
• AI Video Creation: Plan, Generate, and Edit - A beginner introduction to an AI-assisted video-making process, from a simple story idea through generation and editing. The selected videos cover AI animated-story creation, cinematic AI video generation, a beginner AI-video tutorial, and an AI storytelling workflow with CapCut. Tools, availability, and output rights vary; use assets you have permission to use, label synthetic or altered material where appropriate, and review the finished video for errors before sharing.

Frequently Asked Questions

Are there free courses available?

Yes, Starter courses and introductory lessons for all full courses are available as a free preview. No credit card is required to start learning.

How does the 7-day trial work?

Eligible users receive a 7-day free trial. After the trial, your chosen monthly or annual plan renews automatically until you cancel. Cancel before the trial ends to avoid a charge.

What is the Academy Pass?

The Academy Pass grants you unlimited access to all full courses, premium resources, and exclusive assignments. It is available for $14.99/month or $149/year, saving 17% on yearly, and you can cancel anytime through our Stripe billing portal.

Is Academy included with Lumi Pro?

Yes. If you have an active Lumi Pro subscription, unlimited access to the One AI Guide Academy is automatically included at no extra cost.

Open-license course content · CC BY licensed · license verified

Only courses with complete, approved licensing evidence and at least one eligible published lesson are included in this catalogue.