# Klay

> Klay maps technical learning across AI research, system design, machine
> learning mathematics, inference engineering and data engineering. The videos
> a lesson points at are public YouTube videos; what Klay adds is the order, the
> written notes and the decision about which video is worth your evening. The
> opening lessons of each course are open in full; the rest are paid.

## Start here
- [Klay](https://klaylearn.com/): course directory and the main discovery surface.
- [Roadmaps](https://klaylearn.com/roadmap): every module of every track, in order.
- [Pricing](https://klaylearn.com/pricing): the free tier and the one-time lifetime option.

## Courses
- [Everything Data](https://klaylearn.com/everything-data): 12 modules, 85 lessons - the data engineering stack, from ingestion to warehouse.
- [Inference Engineering](https://klaylearn.com/inference-engineering): 8 modules, 94 lessons - how a trained model is actually served - the generation loop, the hardware floor, the KV cache, batching and scheduling, quantisation and speculative decoding, and running a fleet.
- [Deployment](https://klaylearn.com/deployment): 11 modules, 56 lessons - deployment and reliability for someone who has shipped something live - the mathematics of queueing and availability, what deployment and a server actually are, the network from speed of light to DNS and TLS and the edge, processes and containers and serverless, comparing where code runs, state and consistency and consensus, access and blast radius, releasing and rollback, SLOs and alerting and incidents, retry storms and cascading failure and backpressure, and what scale costs.
- [User Systems](https://klaylearn.com/user-systems): 7 modules, 34 lessons - the user layer of a product - identity and authentication, passkeys and sessions, OAuth and OIDC, permissions and fine-grained authorization, teams and multi-tenancy, the account lifecycle and deletion, onboarding and activation, notifications and delivery, support and impersonation and abuse, and what an AI agent may do on a user's behalf.
- [Payments](https://klaylearn.com/payments): 7 modules, 35 lessons - how money actually moves and what has to be true in your database before you call an order paid - authorization and capture and settlement, cards and UPI and bank transfers, gateways and merchants of record, checkout and server-side price authority, minor units and webhooks and idempotency and the double-entry ledger, subscriptions and the RBI e-mandate framework, declines and dunning and chargebacks, marketplaces and payouts, and what spending authority an AI agent may be granted.
- [SEO](https://klaylearn.com/seo): 10 modules, 52 lessons - how anyone finds a product at all, taught from the mechanism up - what a crawler fetches, the inverted index, BM25 and PageRank, then how an AI answer is retrieved, grounded and cited; robots.txt and the AI crawlers, rendering and canonicals, structured data and Core Web Vitals and the page an agent sees, search intent and topic architecture, content with evidence behind it, links and mentions and entities, measurement across Search Console and Bing, and turning a visit into a user.
- [Agents & MCP](https://klaylearn.com/agents-mcp): 13 modules, 60 lessons - the agent loop and everything wrapped around it - the loop itself and its stopping conditions, tools and function calling and schemas, MCP as a protocol with its hosts and clients and servers and transports, building and testing and shipping a server, context engineering from the budget to compaction to memory to context rot, planning and decomposition, multi-agent orchestration and when one agent is the right answer, permissions and sandboxing, prompt injection and the lethal trifecta and the OWASP LLM Top 10, evaluating a trajectory rather than an output, where the tokens go, and tracing a run you have to reproduce.

## Free surfaces
- [Glossary](https://klaylearn.com/glossary): plain definitions of the terms this library teaches, each linked to the lesson that goes further.
- [Engineering levels](https://klaylearn.com/levels): what each rung from L4 to L8 is judged on, plus the Meta, Amazon, Microsoft and Apple ladders.
- [Labs](https://klaylearn.com/labs): small interactive tools that make one idea concrete.
- [Daily papers](https://klaylearn.com/daily): foundational papers with a short reading note each.
- [Tools](https://klaylearn.com/tools): a curated directory, one sentence per tool.

## Every lesson
- https://klaylearn.com/agents-mcp/l/agent-vs-workflow
- https://klaylearn.com/agents-mcp/l/budgets-and-circuit-breakers
- https://klaylearn.com/agents-mcp/l/build-a-real-agent
- https://klaylearn.com/agents-mcp/l/building-a-test-set
- https://klaylearn.com/agents-mcp/l/compaction-and-summarisation
- https://klaylearn.com/agents-mcp/l/context-is-the-budget
- https://klaylearn.com/agents-mcp/l/context-rot
- https://klaylearn.com/agents-mcp/l/credentials-and-secrets
- https://klaylearn.com/agents-mcp/l/defences-that-actually-work
- https://klaylearn.com/agents-mcp/l/designing-a-tool-schema
- https://klaylearn.com/agents-mcp/l/exposing-resources
- https://klaylearn.com/agents-mcp/l/failure-modes-users-see
- https://klaylearn.com/agents-mcp/l/function-calling-basics
- https://klaylearn.com/agents-mcp/l/handoffs-and-routing
- https://klaylearn.com/agents-mcp/l/hosts-clients-servers
- https://klaylearn.com/agents-mcp/l/human-in-the-loop-gates
- https://klaylearn.com/agents-mcp/l/keeping-it-working
- https://klaylearn.com/agents-mcp/l/latency-in-a-loop
- https://klaylearn.com/agents-mcp/l/least-privilege-for-agents
- https://klaylearn.com/agents-mcp/l/llm-as-judge
- https://klaylearn.com/agents-mcp/l/memory-that-is-not-a-vector-db
- https://klaylearn.com/agents-mcp/l/model-routing
- https://klaylearn.com/agents-mcp/l/monitoring-in-production
- https://klaylearn.com/agents-mcp/l/orchestrator-worker
- https://klaylearn.com/agents-mcp/l/owasp-llm-top-10
- https://klaylearn.com/agents-mcp/l/packaging-and-distribution
- https://klaylearn.com/agents-mcp/l/plan-then-execute
- https://klaylearn.com/agents-mcp/l/prompt-caching
- https://klaylearn.com/agents-mcp/l/prompt-injection-with-tools
- https://klaylearn.com/agents-mcp/l/reasoning-and-scratchpads
- https://klaylearn.com/agents-mcp/l/regression-before-deploy
- https://klaylearn.com/agents-mcp/l/reproducing-a-bad-run
- https://klaylearn.com/agents-mcp/l/retrieval-into-the-loop
- https://klaylearn.com/agents-mcp/l/rolling-out-safely
- https://klaylearn.com/agents-mcp/l/sandboxing-execution
- https://klaylearn.com/agents-mcp/l/scoping-what-it-may-do
- https://klaylearn.com/agents-mcp/l/stopping-conditions
- https://klaylearn.com/agents-mcp/l/subagents-and-isolation
- https://klaylearn.com/agents-mcp/l/subtasks-and-checkpoints
- https://klaylearn.com/agents-mcp/l/testing-an-mcp-server
- https://klaylearn.com/agents-mcp/l/the-cost-of-a-committee
- https://klaylearn.com/agents-mcp/l/the-lethal-trifecta
- https://klaylearn.com/agents-mcp/l/the-loop-not-the-model
- https://klaylearn.com/agents-mcp/l/the-mcp-ecosystem
- https://klaylearn.com/agents-mcp/l/too-many-tools
- https://klaylearn.com/agents-mcp/l/tool-errors-and-retries
- https://klaylearn.com/agents-mcp/l/tools-resources-prompts
- https://klaylearn.com/agents-mcp/l/tracing-a-run
- https://klaylearn.com/agents-mcp/l/trajectory-eval
- https://klaylearn.com/agents-mcp/l/transports-stdio-and-http
- https://klaylearn.com/agents-mcp/l/validating-tool-arguments
- https://klaylearn.com/agents-mcp/l/what-mcp-solves
- https://klaylearn.com/agents-mcp/l/what-the-model-actually-sees
- https://klaylearn.com/agents-mcp/l/what-to-log
- https://klaylearn.com/agents-mcp/l/when-more-than-one
- https://klaylearn.com/agents-mcp/l/when-planning-hurts
- https://klaylearn.com/agents-mcp/l/where-the-tokens-go
- https://klaylearn.com/agents-mcp/l/why-output-eval-is-not-enough
- https://klaylearn.com/agents-mcp/l/your-first-agent
- https://klaylearn.com/agents-mcp/l/your-first-mcp-server
- https://klaylearn.com/ai-course/l/agent-failure-modes
- https://klaylearn.com/ai-course/l/ai-in-2026-what-actually-changed
- https://klaylearn.com/ai-course/l/ai-limits-and-integrity
- https://klaylearn.com/ai-course/l/ai-projects-on-your-resume
- https://klaylearn.com/ai-course/l/ai-vs-ml-vs-deep-learning
- https://klaylearn.com/ai-course/l/api-keys-and-secrets
- https://klaylearn.com/ai-course/l/attention-and-transformers
- https://klaylearn.com/ai-course/l/backpropagation-intuition
- https://klaylearn.com/ai-course/l/build-a-doc-qa-app
- https://klaylearn.com/ai-course/l/build-a-neural-net-from-scratch
- https://klaylearn.com/ai-course/l/build-a-streamlit-app
- https://klaylearn.com/ai-course/l/build-gpt-from-scratch
- https://klaylearn.com/ai-course/l/call-an-llm-api-from-python
- https://klaylearn.com/ai-course/l/chunking-documents
- https://klaylearn.com/ai-course/l/contribute-to-open-source
- https://klaylearn.com/ai-course/l/deploy-on-a-free-tier
- https://klaylearn.com/ai-course/l/dot-product-and-similarity
- https://klaylearn.com/ai-course/l/embeddings
- https://klaylearn.com/ai-course/l/fastapi-basics
- https://klaylearn.com/ai-course/l/features-and-labels
- https://klaylearn.com/ai-course/l/functions-and-modules
- https://klaylearn.com/ai-course/l/git-and-github-basics
- https://klaylearn.com/ai-course/l/gradient-descent
- https://klaylearn.com/ai-course/l/how-to-learn-with-ai-not-outsource-to-it
- https://klaylearn.com/ai-course/l/layers-and-activations
- https://klaylearn.com/ai-course/l/lists-dicts-and-loops
- https://klaylearn.com/ai-course/l/loss-functions
- https://klaylearn.com/ai-course/l/matrices-as-transformations
- https://klaylearn.com/ai-course/l/mcp-explained
- https://klaylearn.com/ai-course/l/mean-variance-and-distributions
- https://klaylearn.com/ai-course/l/ml-interview-questions
- https://klaylearn.com/ai-course/l/notebooks-and-virtual-environments
- https://klaylearn.com/ai-course/l/numpy-arrays
- https://klaylearn.com/ai-course/l/overfitting-and-underfitting
- https://klaylearn.com/ai-course/l/pandas-dataframes
- https://klaylearn.com/ai-course/l/pretraining-sft-and-rlhf
- https://klaylearn.com/ai-course/l/probability-intuition
- https://klaylearn.com/ai-course/l/prompt-structure-basics
- https://klaylearn.com/ai-course/l/python-crash-course
- https://klaylearn.com/ai-course/l/pytorch-first-steps
- https://klaylearn.com/ai-course/l/rag-vs-fine-tuning
- https://klaylearn.com/ai-course/l/reading-python-errors
- https://klaylearn.com/ai-course/l/record-a-demo
- https://klaylearn.com/ai-course/l/structured-output-json
- https://klaylearn.com/ai-course/l/supervised-vs-unsupervised
- https://klaylearn.com/ai-course/l/the-agent-loop
- https://klaylearn.com/ai-course/l/tokens-and-tokenization
- https://klaylearn.com/ai-course/l/tokens-context-and-cost
- https://klaylearn.com/ai-course/l/tool-calling
- https://klaylearn.com/ai-course/l/train-test-split
- https://klaylearn.com/ai-course/l/training-vs-inference
- https://klaylearn.com/ai-course/l/vector-search
- https://klaylearn.com/ai-course/l/vectors-and-what-they-mean
- https://klaylearn.com/ai-course/l/what-a-derivative-actually-is
- https://klaylearn.com/ai-course/l/what-is-a-neuron
- https://klaylearn.com/ai-course/l/what-is-an-ai-agent
- https://klaylearn.com/ai-course/l/what-is-machine-learning
- https://klaylearn.com/ai-course/l/what-to-learn-next
- https://klaylearn.com/ai-course/l/why-ai-took-off-now
- https://klaylearn.com/ai-course/l/why-llms-hallucinate
- https://klaylearn.com/ai-course/l/why-rag-exists
- https://klaylearn.com/ai-course/l/write-a-readme-that-gets-read
- https://klaylearn.com/ai-course/l/write-your-first-eval
- https://klaylearn.com/ai-course/l/your-first-sklearn-model
- https://klaylearn.com/deployment/l/access-control-at-data-layer
- https://klaylearn.com/deployment/l/alerting
- https://klaylearn.com/deployment/l/amdahl-and-the-ceiling
- https://klaylearn.com/deployment/l/autoscaling-as-a-control-loop
- https://klaylearn.com/deployment/l/availability-is-multiplication
- https://klaylearn.com/deployment/l/backpressure
- https://klaylearn.com/deployment/l/cascading-failure
- https://klaylearn.com/deployment/l/cdns-and-the-edge
- https://klaylearn.com/deployment/l/choosing-boring-technology
- https://klaylearn.com/deployment/l/connections-cost
- https://klaylearn.com/deployment/l/consensus
- https://klaylearn.com/deployment/l/containers
- https://klaylearn.com/deployment/l/correlated-failure
- https://klaylearn.com/deployment/l/cost-curves-and-crossovers
- https://klaylearn.com/deployment/l/course-opener
- https://klaylearn.com/deployment/l/databases-under-load
- https://klaylearn.com/deployment/l/dns
- https://klaylearn.com/deployment/l/economics-and-when-to-leave
- https://klaylearn.com/deployment/l/feedback-lag-overshoot
- https://klaylearn.com/deployment/l/health-checks-and-draining
- https://klaylearn.com/deployment/l/images-and-reproducibility
- https://klaylearn.com/deployment/l/incident-response
- https://klaylearn.com/deployment/l/least-privilege-blast-radius
- https://klaylearn.com/deployment/l/littles-law
- https://klaylearn.com/deployment/l/logs-metrics-and-traces
- https://klaylearn.com/deployment/l/long-running-servers
- https://klaylearn.com/deployment/l/migrations-against-live-traffic
- https://klaylearn.com/deployment/l/mtbf-mttr-and-their-limits
- https://klaylearn.com/deployment/l/orchestration
- https://klaylearn.com/deployment/l/permissions-as-sets
- https://klaylearn.com/deployment/l/processes-and-scheduling
- https://klaylearn.com/deployment/l/replication-and-consistency
- https://klaylearn.com/deployment/l/retry-storms
- https://klaylearn.com/deployment/l/rollback-beats-forward-fix
- https://klaylearn.com/deployment/l/sampling-and-tails
- https://klaylearn.com/deployment/l/scaling-axes
- https://klaylearn.com/deployment/l/serverless-internals
- https://klaylearn.com/deployment/l/slis-slos-and-error-budgets
- https://klaylearn.com/deployment/l/speed-of-light-is-a-floor
- https://klaylearn.com/deployment/l/static-and-edge
- https://klaylearn.com/deployment/l/strategies-compared
- https://klaylearn.com/deployment/l/the-curve-collected
- https://klaylearn.com/deployment/l/the-four-axes
- https://klaylearn.com/deployment/l/the-hockey-stick
- https://klaylearn.com/deployment/l/the-problem-stated
- https://klaylearn.com/deployment/l/the-release-problem
- https://klaylearn.com/deployment/l/the-trust-boundary
- https://klaylearn.com/deployment/l/tls
- https://klaylearn.com/deployment/l/virtual-machines
- https://klaylearn.com/deployment/l/what-a-server-actually-is
- https://klaylearn.com/deployment/l/what-deployment-means
- https://klaylearn.com/deployment/l/when-machines-make-it-slower
- https://klaylearn.com/deployment/l/why-averages-lie
- https://klaylearn.com/deployment/l/why-local-is-not-production
- https://klaylearn.com/deployment/l/why-state-ruins-everything
- https://klaylearn.com/deployment/l/your-agent-has-the-keys
- https://klaylearn.com/everything-data/l/airflow-dag-authoring
- https://klaylearn.com/everything-data/l/apache-spark-dataframes
- https://klaylearn.com/everything-data/l/backpressure-replay-and-failure-recovery
- https://klaylearn.com/everything-data/l/batch-stream-and-serving-boundaries
- https://klaylearn.com/everything-data/l/bigquery-for-large-scale-analytics
- https://klaylearn.com/everything-data/l/building-training-data-jobs
- https://klaylearn.com/everything-data/l/catalogs-lineage-and-access-control
- https://klaylearn.com/everything-data/l/change-data-capture
- https://klaylearn.com/everything-data/l/contract-tests-for-pipelines
- https://klaylearn.com/everything-data/l/cost-and-resource-management
- https://klaylearn.com/everything-data/l/ctes-for-readable-transformations
- https://klaylearn.com/everything-data/l/dagster-software-defined-assets
- https://klaylearn.com/everything-data/l/data-contracts-and-governance
- https://klaylearn.com/everything-data/l/data-curation-for-foundation-models
- https://klaylearn.com/everything-data/l/data-drift-detection
- https://klaylearn.com/everything-data/l/data-mesh-and-domain-ownership
- https://klaylearn.com/everything-data/l/data-platform-reference-architecture
- https://klaylearn.com/everything-data/l/data-quality-dimensions
- https://klaylearn.com/everything-data/l/databricks-and-unified-analytics
- https://klaylearn.com/everything-data/l/dataset-versioning-principles
- https://klaylearn.com/everything-data/l/dbt-models-for-ml-transformations
- https://klaylearn.com/everything-data/l/deduplication-and-contamination-control
- https://klaylearn.com/everything-data/l/distributed-compute-fundamentals
- https://klaylearn.com/everything-data/l/document-ingestion-and-chunking
- https://klaylearn.com/everything-data/l/dvc-for-data-and-pipelines
- https://klaylearn.com/everything-data/l/embedding-pipeline-design
- https://klaylearn.com/everything-data/l/embeddings-and-vector-search
- https://klaylearn.com/everything-data/l/entities-events-and-observations
- https://klaylearn.com/everything-data/l/experiment-tracking-metadata
- https://klaylearn.com/everything-data/l/feast-feature-store-walkthrough
- https://klaylearn.com/everything-data/l/feature-engineering-lifecycle
- https://klaylearn.com/everything-data/l/feature-reuse-discovery-and-governance
- https://klaylearn.com/everything-data/l/feature-store-data-models
- https://klaylearn.com/everything-data/l/flink-stream-processing
- https://klaylearn.com/everything-data/l/from-data-warehouse-to-ai-platform
- https://klaylearn.com/everything-data/l/great-expectations-validation
- https://klaylearn.com/everything-data/l/iceberg-delta-and-table-formats
- https://klaylearn.com/everything-data/l/incident-response-for-data-pipelines
- https://klaylearn.com/everything-data/l/incremental-batch-processing
- https://klaylearn.com/everything-data/l/instruction-tuning-datasets
- https://klaylearn.com/everything-data/l/kafka-producers-consumers-and-topics
- https://klaylearn.com/everything-data/l/label-and-concept-drift
- https://klaylearn.com/everything-data/l/label-modeling-and-leakage
- https://klaylearn.com/everything-data/l/lakefs-for-data-lake-branches
- https://klaylearn.com/everything-data/l/lakehouse-architecture-overview
- https://klaylearn.com/everything-data/l/llm-data-lifecycle
- https://klaylearn.com/everything-data/l/llm-evaluation-data-management
- https://klaylearn.com/everything-data/l/milvus-and-open-source-vector-search
- https://klaylearn.com/everything-data/l/ml-observability-dashboards
- https://klaylearn.com/everything-data/l/modeling-for-multi-tenant-products
- https://klaylearn.com/everything-data/l/normalization-vs-denormalization
- https://klaylearn.com/everything-data/l/offline-vs-online-features
- https://klaylearn.com/everything-data/l/online-serving-and-low-latency-access
- https://klaylearn.com/everything-data/l/orchestration-for-ml-training
- https://klaylearn.com/everything-data/l/parquet-columnar-storage
- https://klaylearn.com/everything-data/l/pgvector-in-postgres
- https://klaylearn.com/everything-data/l/pinecone-managed-vector-search
- https://klaylearn.com/everything-data/l/pipeline-design-patterns
- https://klaylearn.com/everything-data/l/point-in-time-correct-joins
- https://klaylearn.com/everything-data/l/point-in-time-feature-retrieval
- https://klaylearn.com/everything-data/l/prefect-flows-and-deployments
- https://klaylearn.com/everything-data/l/query-plans-and-optimization
- https://klaylearn.com/everything-data/l/rag-evaluation-and-retrieval-quality
- https://klaylearn.com/everything-data/l/ray-for-data-and-ml-workloads
- https://klaylearn.com/everything-data/l/reproducibility-problem-in-ml
- https://klaylearn.com/everything-data/l/reproducible-training-splits
- https://klaylearn.com/everything-data/l/retries-idempotency-and-backfills
- https://klaylearn.com/everything-data/l/rlhf-and-preference-data
- https://klaylearn.com/everything-data/l/schema-registry-and-event-contracts
- https://klaylearn.com/everything-data/l/slowly-changing-dimensions
- https://klaylearn.com/everything-data/l/snowflake-for-ml-data
- https://klaylearn.com/everything-data/l/spark-performance-tuning
- https://klaylearn.com/everything-data/l/sql-for-training-datasets
- https://klaylearn.com/everything-data/l/sql-testing-and-assertions
- https://klaylearn.com/everything-data/l/star-schemas-for-analytics-and-ml
- https://klaylearn.com/everything-data/l/streaming-features-for-ml
- https://klaylearn.com/everything-data/l/streaming-system-concepts
- https://klaylearn.com/everything-data/l/synthetic-data-generation-pipelines
- https://klaylearn.com/everything-data/l/tecton-and-managed-feature-platforms
- https://klaylearn.com/everything-data/l/tokenization-and-dataset-packing
- https://klaylearn.com/everything-data/l/tooling-landscape-for-ml-engineers
- https://klaylearn.com/everything-data/l/training-serving-data-lineage
- https://klaylearn.com/everything-data/l/warehouse-vs-lakehouse-tradeoffs
- https://klaylearn.com/everything-data/l/window-functions-for-features
- https://klaylearn.com/everything-data/l/workflow-orchestration-fundamentals
- https://klaylearn.com/inference-engineering/l/activation-quantisation
- https://klaylearn.com/inference-engineering/l/agents-per-megawatt
- https://klaylearn.com/inference-engineering/l/arithmetic-intensity
- https://klaylearn.com/inference-engineering/l/autoscaling-and-cold-starts
- https://klaylearn.com/inference-engineering/l/benchmarking-your-server
- https://klaylearn.com/inference-engineering/l/cache-arithmetic
- https://klaylearn.com/inference-engineering/l/cache-aware-routing
- https://klaylearn.com/inference-engineering/l/calibration-and-quality
- https://klaylearn.com/inference-engineering/l/choosing-the-hardware
- https://klaylearn.com/inference-engineering/l/chunked-prefill
- https://klaylearn.com/inference-engineering/l/context-compaction
- https://klaylearn.com/inference-engineering/l/continuous-batching
- https://klaylearn.com/inference-engineering/l/cost-per-million-tokens
- https://klaylearn.com/inference-engineering/l/cost-per-solved-task
- https://klaylearn.com/inference-engineering/l/determinism
- https://klaylearn.com/inference-engineering/l/disaggregation
- https://klaylearn.com/inference-engineering/l/draft-models
- https://klaylearn.com/inference-engineering/l/eagle-and-mtp
- https://klaylearn.com/inference-engineering/l/effort-and-budgets
- https://klaylearn.com/inference-engineering/l/eviction
- https://klaylearn.com/inference-engineering/l/expert-parallelism
- https://klaylearn.com/inference-engineering/l/first-batch
- https://klaylearn.com/inference-engineering/l/flash-attention
- https://klaylearn.com/inference-engineering/l/flops-and-bytes
- https://klaylearn.com/inference-engineering/l/forward-pass-vs-generation
- https://klaylearn.com/inference-engineering/l/fragmentation
- https://klaylearn.com/inference-engineering/l/goodput-and-slos
- https://klaylearn.com/inference-engineering/l/grammar-compilation-cost
- https://klaylearn.com/inference-engineering/l/hit-rate
- https://klaylearn.com/inference-engineering/l/kv-cache-quantisation
- https://klaylearn.com/inference-engineering/l/kv-offload-and-reload
- https://klaylearn.com/inference-engineering/l/launch-overhead
- https://klaylearn.com/inference-engineering/l/logits
- https://klaylearn.com/inference-engineering/l/long-context-prefill
- https://klaylearn.com/inference-engineering/l/mha-mqa-gqa
- https://klaylearn.com/inference-engineering/l/mixture-of-experts
- https://klaylearn.com/inference-engineering/l/moving-the-kv-cache
- https://klaylearn.com/inference-engineering/l/multi-tenancy-and-lora
- https://klaylearn.com/inference-engineering/l/naming-your-bottleneck
- https://klaylearn.com/inference-engineering/l/number-formats
- https://klaylearn.com/inference-engineering/l/one-request-at-a-time
- https://klaylearn.com/inference-engineering/l/paged-attention
- https://klaylearn.com/inference-engineering/l/parallel-sampling
- https://klaylearn.com/inference-engineering/l/pipeline-parallelism
- https://klaylearn.com/inference-engineering/l/precision
- https://klaylearn.com/inference-engineering/l/prefill-and-decode
- https://klaylearn.com/inference-engineering/l/prefill-decode-interference
- https://klaylearn.com/inference-engineering/l/prefix-caching
- https://klaylearn.com/inference-engineering/l/profiling-a-forward-pass
- https://klaylearn.com/inference-engineering/l/radix-attention
- https://klaylearn.com/inference-engineering/l/reading-the-device
- https://klaylearn.com/inference-engineering/l/rollouts-are-inference
- https://klaylearn.com/inference-engineering/l/routing-requests
- https://klaylearn.com/inference-engineering/l/sampling
- https://klaylearn.com/inference-engineering/l/self-host-or-api
- https://klaylearn.com/inference-engineering/l/serving-two-models
- https://klaylearn.com/inference-engineering/l/session-affinity
- https://klaylearn.com/inference-engineering/l/slos-for-thinking
- https://klaylearn.com/inference-engineering/l/speculative-decoding
- https://klaylearn.com/inference-engineering/l/stacking-the-wins
- https://klaylearn.com/inference-engineering/l/static-batching
- https://klaylearn.com/inference-engineering/l/stopping
- https://klaylearn.com/inference-engineering/l/structured-output
- https://klaylearn.com/inference-engineering/l/tensor-parallelism
- https://klaylearn.com/inference-engineering/l/test-time-compute
- https://klaylearn.com/inference-engineering/l/the-agent-workload
- https://klaylearn.com/inference-engineering/l/the-context-budget
- https://klaylearn.com/inference-engineering/l/the-decode-heavy-workload
- https://klaylearn.com/inference-engineering/l/the-generation-loop-lesson
- https://klaylearn.com/inference-engineering/l/the-interconnect
- https://klaylearn.com/inference-engineering/l/the-kv-memory-hierarchy
- https://klaylearn.com/inference-engineering/l/the-memory-hierarchy
- https://klaylearn.com/inference-engineering/l/the-memory-wall
- https://klaylearn.com/inference-engineering/l/the-roofline
- https://klaylearn.com/inference-engineering/l/the-scheduler-loop
- https://klaylearn.com/inference-engineering/l/the-serving-stack
- https://klaylearn.com/inference-engineering/l/the-tool-schema-tax
- https://klaylearn.com/inference-engineering/l/throughput-vs-latency
- https://klaylearn.com/inference-engineering/l/tokenisation-at-serving-time
- https://klaylearn.com/inference-engineering/l/tool-calls-in-the-loop
- https://klaylearn.com/inference-engineering/l/ttft-tpot-itl
- https://klaylearn.com/inference-engineering/l/two-ways-to-go-faster
- https://klaylearn.com/inference-engineering/l/unpredictable-output-length
- https://klaylearn.com/inference-engineering/l/verifiers-in-the-serving-path
- https://klaylearn.com/inference-engineering/l/weight-only-quantisation
- https://klaylearn.com/inference-engineering/l/what-a-reasoning-model-emits
- https://klaylearn.com/inference-engineering/l/what-is-actually-stored
- https://klaylearn.com/inference-engineering/l/when-one-gpu-is-not-enough
- https://klaylearn.com/inference-engineering/l/when-speculation-loses
- https://klaylearn.com/inference-engineering/l/where-the-memory-went
- https://klaylearn.com/inference-engineering/l/why-cache-anything
- https://klaylearn.com/inference-engineering/l/why-decode-is-memory-bound
- https://klaylearn.com/inference-engineering/l/why-the-gpu-waits
- https://klaylearn.com/inference-engineering/l/your-baseline
- https://klaylearn.com/payments/l/3ds-sca-and-the-soft-decline
- https://klaylearn.com/payments/l/agents-that-spend-money
- https://klaylearn.com/payments/l/cards-upi-wallets-and-bank-transfers
- https://klaylearn.com/payments/l/chargebacks-and-the-evidence-that-wins
- https://klaylearn.com/payments/l/dunning-cancellation-and-the-end
- https://klaylearn.com/payments/l/every-state-a-payment-can-be-in
- https://klaylearn.com/payments/l/fraud-card-testing-and-risk-rules
- https://klaylearn.com/payments/l/gateway-processor-or-merchant-of-record
- https://klaylearn.com/payments/l/hosted-embedded-or-your-own-form
- https://klaylearn.com/payments/l/idempotency-or-you-will-charge-twice
- https://klaylearn.com/payments/l/intents-authorization-and-capture
- https://klaylearn.com/payments/l/never-store-money-in-a-float
- https://klaylearn.com/payments/l/payouts-reserves-and-the-float
- https://klaylearn.com/payments/l/recurring-in-india
- https://klaylearn.com/payments/l/refunds-and-what-they-do-not-unwind
- https://klaylearn.com/payments/l/selling-from-india
- https://klaylearn.com/payments/l/selling-to-the-world
- https://klaylearn.com/payments/l/splits-commissions-and-seller-kyc
- https://klaylearn.com/payments/l/test-mode-is-a-different-planet
- https://klaylearn.com/payments/l/the-account-you-cannot-get
- https://klaylearn.com/payments/l/the-amount-must-never-come-from-the-browser
- https://klaylearn.com/payments/l/the-entitlement-state-machine
- https://klaylearn.com/payments/l/the-ledger-and-the-monthly-close
- https://klaylearn.com/payments/l/the-payment-code-your-agent-wrote
- https://klaylearn.com/payments/l/the-tax-you-stop-owing
- https://klaylearn.com/payments/l/the-webhook-is-the-only-thing-that-knows
- https://klaylearn.com/payments/l/trials-coupons-proration-and-upgrades
- https://klaylearn.com/payments/l/usage-billing-credits-and-overages
- https://klaylearn.com/payments/l/what-a-checkout-page-loses
- https://klaylearn.com/payments/l/what-a-payment-costs-you
- https://klaylearn.com/payments/l/what-a-subscription-actually-is
- https://klaylearn.com/payments/l/what-happens-when-someone-pays
- https://klaylearn.com/payments/l/when-the-money-is-not-yours
- https://klaylearn.com/payments/l/who-says-no-and-why
- https://klaylearn.com/payments/l/why-good-customers-get-declined
- https://klaylearn.com/seo/l/activation
- https://klaylearn.com/seo/l/ai-visibility-tools
- https://klaylearn.com/seo/l/attribution-and-experiments
- https://klaylearn.com/seo/l/bing-webmaster-and-copilot
- https://klaylearn.com/seo/l/brand-demand
- https://klaylearn.com/seo/l/canonicals-and-the-url-contract
- https://klaylearn.com/seo/l/core-web-vitals
- https://klaylearn.com/seo/l/decay-refresh-and-consolidation
- https://klaylearn.com/seo/l/digital-pr-and-the-earned-mention
- https://klaylearn.com/seo/l/eeat-as-evidence
- https://klaylearn.com/seo/l/embeddings
- https://klaylearn.com/seo/l/field-vs-lab-data
- https://klaylearn.com/seo/l/instrumenting-the-funnel
- https://klaylearn.com/seo/l/javascript-seo
- https://klaylearn.com/seo/l/keyword-research-when-the-query-is-a-conversation
- https://klaylearn.com/seo/l/links-as-votes
- https://klaylearn.com/seo/l/original-research-as-a-moat
- https://klaylearn.com/seo/l/pricing-free-tiers-and-trials
- https://klaylearn.com/seo/l/product-led-sales-and-expansion
- https://klaylearn.com/seo/l/programmatic-seo
- https://klaylearn.com/seo/l/query-fan-out
- https://klaylearn.com/seo/l/reading-the-serp
- https://klaylearn.com/seo/l/referral-and-viral-loops
- https://klaylearn.com/seo/l/rerankers-and-citations
- https://klaylearn.com/seo/l/retention-and-cohorts
- https://klaylearn.com/seo/l/retrieval-augmented-generation
- https://klaylearn.com/seo/l/robots-status-codes-noindex
- https://klaylearn.com/seo/l/scaled-content-abuse
- https://klaylearn.com/seo/l/search-console-honestly
- https://klaylearn.com/seo/l/semantic-html-and-headings
- https://klaylearn.com/seo/l/site-architecture-and-internal-links
- https://klaylearn.com/seo/l/sitemaps-indexnow-and-bing
- https://klaylearn.com/seo/l/structured-data-in-2026
- https://klaylearn.com/seo/l/tf-idf-and-bm25
- https://klaylearn.com/seo/l/the-ai-crawler-landscape
- https://klaylearn.com/seo/l/the-answer-engines
- https://klaylearn.com/seo/l/the-answer-first-landing-page
- https://klaylearn.com/seo/l/the-crawler
- https://klaylearn.com/seo/l/the-discovery-operating-system
- https://klaylearn.com/seo/l/the-entity
- https://klaylearn.com/seo/l/the-generative-ai-reports
- https://klaylearn.com/seo/l/the-inverted-index
- https://klaylearn.com/seo/l/the-landing-seam
- https://klaylearn.com/seo/l/the-off-site-graph
- https://klaylearn.com/seo/l/the-page-an-agent-sees
- https://klaylearn.com/seo/l/the-page-that-answers-first
- https://klaylearn.com/seo/l/time-to-value
- https://klaylearn.com/seo/l/topic-clusters-and-entities
- https://klaylearn.com/seo/l/what-a-query-actually-is
- https://klaylearn.com/seo/l/what-links-still-buy
- https://klaylearn.com/seo/l/what-people-first-actually-means
- https://klaylearn.com/seo/l/why-the-same-prompt-answers-twice
- https://klaylearn.com/user-systems/l/abuse-moderation-and-account-safety
- https://klaylearn.com/user-systems/l/account-changes-merging-and-recovery
- https://klaylearn.com/user-systems/l/activation-states-and-re-engagement
- https://klaylearn.com/user-systems/l/admin-panels-and-user-operations
- https://klaylearn.com/user-systems/l/agents-acting-on-behalf-of-users
- https://klaylearn.com/user-systems/l/ai-support-agents-and-human-handoff
- https://klaylearn.com/user-systems/l/authentication-is-not-authorization
- https://klaylearn.com/user-systems/l/building-support-into-the-product
- https://klaylearn.com/user-systems/l/delivery-retries-and-provider-failures
- https://klaylearn.com/user-systems/l/designing-the-first-value-journey
- https://klaylearn.com/user-systems/l/directing-and-auditing-an-agent-that-builds-auth
- https://klaylearn.com/user-systems/l/email-push-and-in-app
- https://klaylearn.com/user-systems/l/email-verification-and-account-recovery
- https://klaylearn.com/user-systems/l/fine-grained-authorization-and-rebac
- https://klaylearn.com/user-systems/l/impersonation-audit-logs-and-privileged-access
- https://klaylearn.com/user-systems/l/invitations-scim-and-enterprise-lifecycle
- https://klaylearn.com/user-systems/l/oauth-and-openid-connect
- https://klaylearn.com/user-systems/l/passkeys-and-phishing-resistant-auth
- https://klaylearn.com/user-systems/l/passwords-otp-and-magic-links
- https://klaylearn.com/user-systems/l/preferences-frequency-and-state
- https://klaylearn.com/user-systems/l/privacy-data-export-and-deletion
- https://klaylearn.com/user-systems/l/profiles-preferences-and-user-data
- https://klaylearn.com/user-systems/l/progressive-onboarding-and-invitations
- https://klaylearn.com/user-systems/l/security-boundaries-and-failure-modes
- https://klaylearn.com/user-systems/l/sessions-cookies-and-stolen-tokens
- https://klaylearn.com/user-systems/l/signup-is-not-activation
- https://klaylearn.com/user-systems/l/teams-organizations-and-multi-tenancy
- https://klaylearn.com/user-systems/l/the-architecture-of-a-user-system
- https://klaylearn.com/user-systems/l/the-idor-in-your-api
- https://klaylearn.com/user-systems/l/the-notification-system
- https://klaylearn.com/user-systems/l/the-user-lifecycle-state-machine
- https://klaylearn.com/user-systems/l/transactional-vs-marketing
- https://klaylearn.com/user-systems/l/what-a-user-actually-is
- https://klaylearn.com/user-systems/l/when-the-user-system-breaks

## Optional
- [Answer reference](https://klaylearn.com/llms-reference.txt): the full free text of the glossary and levels pages, in one file.
- [Page index](https://klaylearn.com/llms-full.txt): every URL with the description it ships.
- [XML sitemap](https://klaylearn.com/sitemap.xml): 1100 canonical URLs.