Skip to content
anythingai.space
The illustrated course

Every idea, drawn and set in motion.

Each lesson is a diagram you step through. Watch requests flow, caches fill, agents loop and things break — then see the fix land. Go in order or jump to what you need.

25 modules268 lessons2409 animated steps55h of material
tokenize“I LOVE THIS PRODUCT!” → TOKENS (WITH A 2-D SLICE OF EACH EMBEDDING)I[0.1, 0.4]love[0.9, 0.8]this[0.2, 0.1]product[0.5, 0.3]![0.6, 0.7]Same pipeline, different heads: translate · classify · answer · summarise ·generateReview textEmbeddingstoken → vectorTransformerreads in contextPositivesentiment 0.9601

AI Foundations

Machine learning, deep learning, NLP, computer vision, reinforcement learning and the techniques underneath modern AI.

34 lessons427 min0/34
ENCODER · SEES EVERYTHINGThecatsatonmatThecatsatonmatDECODER · SEES ONLY THE PASTThecatsatonmatThecatsatonmatBERT-style → intent, sentiment,embedding modelsGPT-style → text generation, chat,agents02

How LLMs Work

Tokens, attention, transformers, embeddings and the inference pipeline — from text in to text out.

6 lessons74 min0/6
FRONTIERWORKHORSEFAST / EFFICIENTSMALL / EDGEFable 5.1$10 / $50GPT-6 Astra$10 / $50Opus 5.5$4 / $20Sonnet 5.5$2 / $10GPT-6.1 Sol$2 / $10GPT-6.1 Terra$1 / $5GPT-6 Luna$0.10 / $0.50Haiku 5.5$0.10 / $0.50 <100KGemini 3.8 Flash$0.75 / $3.75 introGemma 4 12BApache 2.0Qwen3.8-27BApache 2.0 · denseGranite 4.23B / 8B · Apache03

Choosing Models

The model landscape, capability assessment, pricing, and how to pick the right model for the job.

4 lessons49 min0/4
FSDP / ZERO: STATE SHARDED ACROSS GPUSPER PARAMETER ≈ 16 BYTES (TYPICAL SPLIT)BF16 weights · 2 BBF16 gradients · 2 BFP32 master copy · 4 BAdam momentum · 4 BAdam variance · 4 BGPU 11/4 of the stateGPU 21/4 of the stateGPU 31/4 of the stateGPU 41/4 of the state04

Training & Adaptation

Pretraining, fine-tuning, LoRA, RLHF/DPO, distillation, synthetic data, quantization and reasoning models.

8 lessons103 min0/8
COMPUTE PER BYTE OF MEMORY BANDWIDTHH100 · BF16~300 FLOPs/byteVera Rubin · NVFP4~2,600 FLOPs/byteH100: ~1 PFLOPS dense BF16 ÷ 3.35 TB/sHBM3Rubin: ~50 PFLOPS NVFP4 ÷ 19.2 TB/s HBM4Decode reads the weights once per step — so batch many sequences(or verify several speculative tokens) to use the compute you paidfor05

Inference & Serving

KV cache, batching, speculative decoding, paged attention, serving stacks and the cost playbook.

9 lessons112 min0/9
CONSISTENCY AS A SYSTEM PROPERTYtemperature=0Effort is the dial now,not sampling.Eval harnesstemperature: 0Model APIversion 2026-10-21400Pin model+effortSchema / enumInput-hash cachePass rateN runs per case06

Prompting & Context

Prompt fundamentals, few-shot, chain/tree of thought, context engineering, structured output and injection defense.

8 lessons90 min0/8
OFFLINE · INDEXINGONE DOC → CHUNKS§1 Eligibility§2 Refund window: 14 days§3 Annual plans§4 ExceptionsPolicy docs1,200 filesChunker~400 tokensEmbedderVector DB48k chunks07

Retrieval & RAG

Chunking, embeddings, vector databases, hybrid search, reranking, GraphRAG, agentic RAG and RAG at scale.

37 lessons444 min0/37
pinnedverifyTHE EFFORT DIAL (LOW → MAX; GPT-6 ASTRA EXPOSES FIVE LEVELS)lowmedium ← Opus 5.5 defaulthigh ← Opus 5 defaultmaxUnpinned upgrade =silent latency & qualitychangeAgent configmodel + effortReasoning modelthe executiveeffort?Evalsre-run on upgrade08

Agentic Systems

Agent loops, ReAct, tool use and MCP, multi-agent orchestration, planning, recovery, HITL and durable execution.

15 lessons187 min0/15
CONTEXT, EVERY TURNsystem ruleslast month's chatsold tool outputspasted docscurrent question1MINPUT COST PER TURN · OPUS 5.51M uncached$4.001M fully cached$0.2010K retrieved$0.04Plus: attention dilution and slowertime-to-first-token09

Memory & State

Short- and long-term memory, agentic memory, semantic caching and state management.

6 lessons69 min0/6
LANGCHAIN PACKAGE TIERSlangchain-community · long taillangchain-classic · legacy v0partners · -openai, -anthropiclangchain · create_agent, MCPlanggraph · orchestrationlangchain-core · RunnablesLibrary / SDK code → langchain-coreonlyServices → core + partner packages (+langgraph)Notebooks → full langchain forconvenience10

Frameworks & Tools

LangChain, LangGraph, LlamaIndex, DSPy, multi-agent frameworks, coding agents — and how to choose.

6 lessons75 min0/6
passfailPage routertext layer? layout?Flashnative text, no modelsBasicsmall OCR models · CPUStandardparsing VLM (default)Validatetables · order · blanksBlocks + locatorsEscalateadvanced / frontier VLM11

Document Processing

OCR, layout analysis and turning messy documents into clean, retrievable structure.

1 lessons13 min0/1
STAGE 1 · PRE-TRAINING (HUGE, ONCE)STAGE 2 · FINE-TUNING (SMALL, PER TASK)next tokenweightsUnlabeled dataweb · Wikipedia · codeFoundation modelgeneral knowledgeLabeled datatask examplesFine-tuneMedical modelCode model12

Generative AI

Text, image, video, audio and code generation — the model families, design patterns and how they are built.

8 lessons93 min0/8
natural languagestructured callreplyONE TURN CARRIES A GOAL AND ITS DETAILSI needa hotelgoalinPariscitynext weekdatesunderstand → keep track → respondUserAssistantBooking APIbook_hotel(...)13

Conversational AI

Dialogue systems from classical NLU pipelines to LLM assistants and voice — memory, design patterns and evaluation.

8 lessons97 min0/8
POST/v1/conversations201 Created + LocationPOST/v1/conversations/{id}/messages200 text/event-streamGET/v1/conversations/{id}200 JSONGET/v1/kb/articles?q=vpn&cursor=…200 + next_cursorPATCH/v1/tickets/{number}Idempotency-Key200 ticketDELETE/v1/conversations/{id}204 No Content14

Backend for AI

REST, HTTP, async, streaming, auth, databases, caching, queues, jobs, rate limits, idempotency, retries and observability.

20 lessons219 min0/20
NAIVE: CODE COPIED BEFORE DEPSpip install (all deps) — rebuiltCOPY . . — changedpython:3.12 full base + dev toolsOPTIMIZED: DEPS FIRST, CODE LASTCOPY app/ — rebuilt (small)pip install — cachedpython:3.12-slim — cached1-line change → reinstall everything1-line change → one smalllayer15

Deployment & Delivery

Docker, Kubernetes, CI/CD, eval gates, IaC, GPUs, autoscaling, canaries, flags, versioning, health checks and rollbacks.

20 lessons221 min0/20
DAILY BILL, 1M QUERIES × (2K IN + 2K OUT)All on mid-tier~$24K/day65% routed to small~$9K/dayMID-TIER BILL SPLITinput $4Koutput $20KOutput tokens cost 5× input here: cap output length next16

LLM Infrastructure & FinOps

LLM infrastructure, CI/CD for AI, gateways and model routing, FinOps and token economics.

4 lessons53 min0/4
RAW ROWage: 34country: “DE ”income: nullorders: 12duplicate of #882TRAINING-READY FEATURESage_z: 0.21country_DE: 1income: medianorders_per_month: 1.0SPLIT (EXAMPLE PROPORTIONS)train 70%validation 15%test 15%Raw dataCleanTransformFeaturesSplit17

MLOps

The ML lifecycle in production: versioning, registries, CI/CD, infrastructure, automation, monitoring and drift.

7 lessons86 min0/7
ACTIONSlogsmetricseventsWeb tierDatabasesNetworkAIOps platformML on ops dataDetect anomaliesCorrelate eventsPredict issuesAuto-remediate18

AIOps

AI for IT operations: data sources, anomaly detection, correlation, root-cause analysis, tools and use cases.

5 lessons63 min0/5
YOUR AI STACKPrompt + modelGateway (LiteLLM)Serving engine (vLLM, SGLang)Tool servers (MCP SDKs)Vector store (Milvus, pgvector)Unauthenticated RCE in a gatewaythat holds every keyOne bad request killed the enginefor every tenantShared transport sent one client'sresponses to anotherPatch vector stores like databases,not libraries19

AI Security

Prompt injection, jailbreaks, OWASP LLM Top 10, guardrails, PII, least privilege, sandboxing, secrets, approvals, audit and red teaming.

22 lessons269 min0/22
BEFORERefundorder8812,emailuser@example.comcard4111-1111-1111-1111AFTER PII GUARDRefundorder8812,email[EMAIL_REDACTED]card[CREDIT_CARD_REDACTED]UserTopicPIIInjectionLimits20

Reliability & Safety

Guardrails, ensembles, reliability patterns, governance and compliance.

4 lessons52 min0/4
1–5%FRONTIER JUDGE, $/DAY AT 100K TRACESOpus 5.5$1,800GPT-6 Astra$4,500JUDGE LATENCY (VENDOR-REPORTED)Luna-2152 msGPT-4o judge3,200 msTraffic100K traces/dayDistilled judge100% of tracesFrontier judge1–5% sampleHuman reviewdisagreements21

Evaluation & Observability

LLM evaluation, LLM-as-judge, tracing and observability, and reading benchmarks honestly.

5 lessons64 min0/5
NAIVE RAG · MVPSADVANCED RAG · PRODUCTIONstuff top-kQueryVector searchtop-kGenerateRewrite queryHybrid search50 vector + 50 BM25RerankRRF top 20 → best 5+100–200 msGenerate22

Design Patterns

Reusable AI system patterns — and the anti-patterns that sink projects.

2 lessons30 min0/2
CLAUDE CODE INPUT : OUTPUT TOKENSearlier189:1Sep 2026324:1ENTERPRISE APPS WITH AI AGENTS (GARTNER)early 2025< 5%end 2026 (proj.)40%MCP: 10,000+ activeserversLocalCloudIDEVision/GUIManaged23

Tool Use & Computer Agents

Tool-use architectures, computer-use agents, building tool agents safely.

7 lessons92 min0/7
audiotexttokensaudio chunksCallermic audioStreaming STT+ VADTurn detectorText LLM+ toolsStreaming TTSAgent audio24

Voice & Multimodal

Real-time voice agents and multimodal generation.

2 lessons26 min0/2
ONLINE · QUERY PATHDATA LAYER · INGESTIONhit?Employees5,000 · 500 peakAPI gatewayauth · rate limitRBACQuery serviceorchestration4 replicasSonnet 5.5private endpointAnswer cacheexact + semantic30% hitsACL metadataPostgres · RBACHybrid retrieverRRF + ACL filterRerankertop 50 → 10Audit logevery queryQdrant4 shards × 2HAStreaming ingestfresh < 1 hElasticsearch3-node cluster25

Case Studies

Whole systems, end to end: enterprise RAG, coding agents, support automation, fraud, voice AI and more.

20 lessons267 min0/20

Lessons are adapted from the open-source AI System Design Guide (MIT) — the full chapters live in the guide library.