System Designs

Click a layer. See why it exists.

Interactive paths — each flow has its own diagram and its own code. Write paths, deploy on AWS, a request’s blast radius across services, interview walks, language trees you can point at, then AI curricula: fundamentals (ML through domains), advanced generative (LLMs through agents), and the GOD map that is both on one board.

01 · Go

Interactive

Write path · Write-heavy API

Scalable & Low-Latency Go System Design

Invoice creation — sync until commit, then fan out

A create-invoice request that stays fast: validate, authorize, lock, commit — then PDF, email, and analytics leave the hot path.

  • Goroutines
  • ACID
  • Outbox
  • Queues
  • Redis
  • Horizontal scaling

Open flow →

02 · Go

Interactive

Deploy path · Ship & run

Go on AWS: Docker, CI/CD, Lambda

One image → ECR → HTTP + worker Lambdas · two sites on the edge

A Go module becomes one Docker image. CI runs tests and checks, pushes to ECR, then two Lambdas pick it up — cmd/api for HTTP, cmd/worker for the queue. Route 53 and CloudFront put the platform, the client site, and the API on real hostnames.

  • Docker
  • CI/CD
  • ECR
  • Lambda
  • SQS
  • Redis
  • Replica
  • Route 53
  • CloudFront
  • S3

Open flow →

03 · AWS

Interactive

Infra path · Blast radius per module

AWS: a request, the services it touches

Thirty-six layers — click a feature, see which boxes run, and which must not

A user intent is not “the server.” Homepage is the edge. Create is gateway, Lambda, Redis, RDS, then SQS send. PDF and email run on the worker after 201. Click a module; the highlighted services are the count.

  • Route 53
  • CloudFront
  • API Gateway
  • Lambda
  • RDS
  • Redis
  • SQS
  • S3
  • SES
  • IAM

Open flow →

04 · Go

Interactive

Scale path · Scale & isolation

Scale and dependency failures

10,000 RPS · peak latency · DB saturation · slow payment

Design for 10k RPS, then diagnose peak latency. Identify the bottleneck before you pick a lever — indexes and cache, not a bigger pool; timeouts and a breaker, not hope.

  • Horizontal scaling
  • p99
  • Caching
  • Read replicas
  • Timeouts
  • Circuit breaker

Open flow →

05 · Go

Interactive

Correctness path · Invariants

Concurrency and correctness

Duplicate payments · booking · the last item

The same payment twice, many users on one seat, two buyers and one SKU. Uniqueness, TTL holds, and a single UPDATE … WHERE quantity >= 1.

  • Idempotency
  • Reservations
  • Atomic updates
  • Optimistic locking
  • Last item

Open flow →

06 · Go

Interactive

Jobs path · Runtime

Jobs and Go runtime safety

A million jobs · memory · unbounded goroutines

Durable queue, bounded pool, poison to the DLQ. Investigate rising RSS with profiles, not GC. Extra goroutines need a ceiling around scarce resources.

  • Worker pool
  • DLQ
  • pprof
  • Backpressure
  • Graceful shutdown

Open flow →

07 · Go

Interactive

Evolution path · Change

System evolution and deployment

Strangler · zero downtime · expand-and-contract

Extract a module behind a contract, route a slice, keep rollback. Deploy new Go instances with readiness, drain, and graceful shutdown — schema expands before it contracts.

  • Strangler
  • Outbox
  • Rolling
  • Blue-green
  • Canary
  • Expand-contract

Open flow →

08 · Go

Interactive

Language path · Language as a tree

Go: types, concurrency, and the runtime

Forty-nine layers — a package, a goroutine, then channels and the GMP pipeline

A package becomes a process, then the process’s memory: types and methods, slices and maps, error values, then goroutines and channels, then the scheduler — G, M, P — with the traps crossed out.

  • Types
  • Slices
  • Interfaces
  • Errors
  • Goroutines
  • Channels
  • Context
  • GMP
  • Escape analysis
  • GC

Open flow →

09 · Python

Interactive

Language path · Language as a tree

Python: types, asyncio, and the runtime

Forty-nine layers — a module, a Task, then queues and the GIL pipeline

A module becomes a process, then the process’s memory: names and objects, lists and dicts, exceptions, then Tasks and queues, then the loop — Task, thread, GIL — with the traps crossed out.

  • Names
  • Lists
  • Protocols
  • Exceptions
  • asyncio
  • Queues
  • Cancel
  • GIL
  • Refcount
  • GC

Open flow →

10 · React

Interactive

Composition path · UI as a tree

React: props, composition, hooks, and lifecycle

Fifty-six layers — a function, a tree, then hooks and the mount pipeline

A function becomes a tree, then the tree’s memory: props and composition, useState, then the rest of the hooks, then mount/update/unmount — function pipeline on the left, class methods on the right.

  • Props
  • Composition
  • useState
  • useEffect
  • useRef
  • useMemo
  • useContext
  • Custom hooks
  • Function lifecycle
  • Class lifecycle

Open flow →

11 · AI

Interactive

Learning path · Curriculum as a tree

AI fundamentals: ML, deep learning, domains

Sixteen layers — the umbrella, three ML signals, then CNN · RNN · transformer, then NLP · vision · speech · rank

Artificial intelligence reads left to right: how you get a signal, how deep the net is, which sensor you are on, then the job — data, features, train, evaluate, serve — with the magic-and-no-split traps crossed out.

  • Supervised
  • Unsupervised
  • RL
  • CNN
  • RNN / LSTM
  • Transformers
  • NLP
  • Computer vision
  • Speech
  • Recsys

Open flow →

12 · AI

Interactive

Learning path · Curriculum as a tree

AI advanced: generative, RAG, agents

Twenty-four layers — tokens through the window, then tools, then retrieval, then adapters, then the application

Generative AI samples; it is not a database. Tokens through the window, then schema and tools, then RAG beside LoRA. The product is a horizontal application lane: agents, memory, MCP, guardrails, eval, and traces.

  • LLMs
  • Tokens
  • Embeddings
  • Attention
  • RAG
  • Tool calling
  • LoRA
  • Agents
  • MCP
  • Eval

Open flow →

13 · AI

Interactive

Learning path · Curriculum as a tree

AI GOD flow: the whole map

Thirty-nine layers — ML through serve, then LLMs through observability, two trap bands

The full map on one board, each topic a left-to-right lane. Fundamentals on top, generative below. Transformer the architecture and Transformers the LLM are the same block at two altitudes. Click Evaluate, RAG, or Agents.

  • Machine learning
  • Deep learning
  • Domains
  • LLMs
  • RAG
  • Fine-tuning
  • Agents
  • MCP
  • Guardrails
  • Observability

Open flow →