CAPSTONE 01 · APPLICATIONS TRACK
Architect a globally distributed flash-sale booking platform sustaining 500k requests/second: virtual waiting rooms, Redis Lua atomic reservation, and double-entry ledger settlement.
Acceptance SLOs: 100,000 RPS peak burst · p99 < 85ms · Zero inventory drift
Token-bucket & sliding-window edge rate limiting for 50x flash-sale traffic bursts Atomic seat reservation state machine with bounded lease expiration & idempotency keys Transactional outbox + CDC settlement pipeline with zero double-charging or overselling CAPSTONE 02 · INFRASTRUCTURE TRACK
Architect and implement a production-grade distributed key-value store from scratch: LSM storage engine, Raft state machine replication, and consistent hash routing.
Acceptance SLOs: Linearizable Quorum Reads/Writes · < 1.2s leader failover · Zero committed data loss
Append-only WAL + MemTable + SSTable storage engine with Bloom filter point lookups Multi-Raft consensus replication group with leader election, log compaction & split-brain fencing Consistent hashing partition placement with dynamic range splits and Phi accrual failure detection CAPSTONE 03 · DISTRIBUTED AI SERVING
Architect an enterprise-scale distributed LLM inference engine: continuous iteration-level batching, PagedAttention KV cache virtualization, tensor/pipeline parallelism, and speculative decoding.
Acceptance SLOs: TTFT p95 < 180ms · ITL p95 < 22ms/token · > 58% GPU MFU
Iteration-level continuous batching scheduler with prefill/decode disaggregation PagedAttention virtual KV-cache block table manager eliminating external fragmentation (< 4% waste) Prefix caching & speculative decoding pipeline with graceful token-level backpressure CAPSTONE 04 · ENTERPRISE RAG & AGENTIC PLATFORM
Architect an enterprise-scale distributed retrieval-augmented generation (RAG) platform: hybrid dense-sparse search, Reciprocal Rank Fusion, graph knowledge integration, and multi-vector reranking.
Acceptance SLOs: Recall@10 >= 0.94 · End-to-end retrieval p99 < 65ms · Zero cross-tenant leakage
Streaming document ingestion + chunking + hybrid dense (HNSW) and sparse (BM25) retrieval Reciprocal Rank Fusion (RRF) + cross-encoder reranking with strict tenant ACL pre-filtering Durable agentic execution DAG with semantic caching, tool-call sandboxing & citation grounding 18 Systems DocumentedOpen Source (OSI) Source-Available
All Systems (18) Distributed DBs (5) Messaging & Streams (3) Consensus (2) Storage & Filesystems (3) Edge & Ingress (2) AI Platforms (3)
01
Apache Kafka Apache 2.0 Distributed Messaging & Event Logs High-throughput distributed commit log architecture, zero-copy network transfer, and partition replication.
02
Apache Cassandra Apache 2.0 Distributed NoSQL Database Decentralized peer-to-peer wide-column store using Dynamo ring topology and LSM-tree storage.
03
TiDB / TiKV Apache 2.0 Distributed SQL & Multi-Raft KV Horizontally scalable distributed SQL with Multi-Raft consensus, automated range splitting, and Percolator 2PC transactions.
04
CockroachDB BSL 1.1 / CCL Source-Available Distributed SQL & Global Consensus Serializable distributed SQL using Multi-Raft consensus, range leaseholders, and Hybrid Logical Clocks (HLC).
05
etcd Apache 2.0 Consensus & Coordination Strongly consistent distributed key-value store using Raft consensus, bbolt storage, and lease-backed coordination.
06
Apache Flink Apache 2.0 Distributed Stream Processing Low-latency stateful stream processing with event-time watermarking and asynchronous Chandy-Lamport barrier checkpointing.
07
Apache Pulsar & BookKeeper Apache 2.0 Tiered Log & Messaging Multi-tenant publish-subscribe architecture decoupling stateless brokers from stateful BookKeeper storage ledgers.
08
Apache ZooKeeper & Ratis Apache 2.0 Distributed Coordination Hierarchical in-memory znode coordination service implementing Zab atomic broadcast for configuration and leader election.
09
Vitess & ClickHouse Apache 2.0 Sharding Proxy & Columnar OLAP Horizontal MySQL sharding proxy engine (VTGate/VTTablet) combined with ClickHouse vectorized columnar execution.
10
RocksDB Apache 2.0 / GPL 2.0 Embedded LSM-Tree Storage Engine High-performance embedded key-value storage engine built on Write-Ahead Logs, SkipList MemTables, immutable SSTables, and Leveled Compaction.
11
Ceph Storage LGPL 2.1 / GPL 2.0 Unified Object, Block & File Pseudo-random algorithmic object placement via CRUSH maps, RADOS peering, and BlueStore NVMe direct I/O.
12
Apache HDFS & HBase Apache 2.0 Distributed Filesystem & Wide-Column Scale-out distributed append-only storage with NameNode block metadata and HBase region-server LSM engines.
13
Envoy Proxy & Istio Apache 2.0 Service Mesh & Ingress Gateway High-performance C++ edge/service proxy featuring dynamic xDS configuration, circuit breaking, and mTLS sidecar injection.
14
NGINX, HAProxy, Kong & APISIX BSD / Apache 2.0 High-Performance Reverse Proxies Event-driven I/O multiplexing architectures for TLS termination, upstream keepalive pooling, and rate limiting.
15
vLLM Apache 2.0 LLM Inference & Serving High-throughput LLM serving engine featuring PagedAttention virtual memory management for dynamic KV caches.
16
Ray Apache 2.0 Distributed AI Compute Unified distributed computing framework for scaling Python, actor-based stateful workloads, and distributed training.
17
PyTorch Distributed Modified BSD Distributed ML Training Deep learning communication runtime implementing DistributedDataParallel (DDP), FSDP, and tensor parallel collectives.
18
Milvus Apache 2.0 Distributed Vector Database Cloud-native vector database decoupling storage from compute with HNSW, IVF, and scalar filtering.