Resume

Timeline

The 26-week plan, one fundamentals topic and one system design problem a week.

week 1 of 26 · 0 done

block 1 One machine, then the network

  1. w01OS: processes, threads, virtual memory, syscallsDelivery framework, how SD is gradedthis week
  2. w02Concurrency: locks, deadlock, producer-consumerCore concepts: scaling, consistency, CAPupcoming
  3. w03Networking: DNS, TCP, TLS, HTTP/1.1 vs 2, L4 vs L7 LBsCore concepts: networking, APIs, load balancingupcoming
  4. w04API design: REST vs gRPC, pagination, idempotency, versioningURL shortenerupcoming

block 2 Storing data

  1. w05Databases I: indexes, B-tree vs LSM, query plansTicketmasterupcoming
  2. w06Databases II: transactions, isolation, lockingDropboxupcoming
  3. w07Caching: Redis, eviction, invalidation, stampedesNews feedupcoming
  4. w08Streaming: Kafka partitions, consumer groups, delivery semanticsRate limiterupcoming

block 3 Staying up

  1. w09Replication and consensus: quorums, RaftWhatsAppupcoming
  2. w10Resilience: rate limiting, backpressure, retries, LB algorithmsAd click aggregatorupcoming

block 4 How LLMs work

  1. w11LLM basics: tokens, attention, samplingUberupcoming
  2. w12AI harnesses: agent loop, tool calling, context managementWeb crawlerupcoming
  3. w13Evals: measuring LLM systemsTop-Kupcoming
  4. w14RAG: embeddings, vector indexes (HNSW)Distributed job schedulerupcoming

block 5 Serving foundations

  1. w15GPU basics: memory hierarchy, bandwidth, why batching winsPayment systemupcoming
  2. w16Inference: prefill vs decode, KV cache, memory mathYouTubeupcoming
  3. w17Batching: continuous batching, PagedAttentionGoogle Docsupcoming
  4. w18Serving stacks: vLLM, SGLang, TritonLeaderboardupcoming
  5. w19Serving metrics: TTFT, TPOT, p99, SLOsML system design introupcoming

block 6 Serving at scale

  1. w20Quantization: INT8, FP8, 4-bitSearchupcoming
  2. w21Routing: prefix-cache-aware load balancingNotification serviceupcoming
  3. w22Parallelism: tensor, pipeline, expert (MoE)ML system design: recommendationsupcoming
  4. w23Speculative decodingLLM inference APIupcoming
  5. w24Autoscaling: Kubernetes, GPU pools, cold startsDistributed cacheupcoming
  6. w25CUDA basics: kernels, memory hierarchyCold problem, timedupcoming
  7. w26Capstone: LLM serving platform end to endThe capstone itselfupcoming