Architecture

Lightweight Architecture Decision Records

Peter Evans' lightweight ADR repo is a minimal five-section template for capturing architecture decisions in plain markdown files stored alongside code. The approach lowers the adoption barrier enough that even skeptical teams can accept the overhead, keeping decisions documented at the moment they're made.

Architectural Decision Records (ADRs)

The ADR (Architectural Decision Records) site is the canonical resource for documenting significant design choices alongside their context and rationale. It matters because decision records address the institutional memory problem in software teams—why something was built a certain way is often lost over time, leaving future developers unable to distinguish intentional choices from accidents.

Data Engineering Design Patterns (DEDP)

Data Engineering Design Patterns (DEDP) is a free online book covering canonical patterns for building data pipelines — ingestion, transformation, storage, and orchestration. Structured as a pattern catalog rather than a tutorial, useful as a reference for recurring architectural decisions.

Lato: Python Microframework for Modular Monoliths

Lato is a Python microframework for building modular monoliths and loosely coupled applications — explicit module boundaries, dependency injection, and event-driven communication within a single process. A structured alternative to ad-hoc Django app organization.

The Architecture of Today's LLM Applications

GitHub's overview of LLM application architecture patterns as of late 2023 — the standard components (inference, context, orchestration, state, tools) and how they fit together. A useful snapshot of the consensus architecture before it fragmented into frameworks.

Adding Elixir to a Next.js Chat App

Linen's engineering post on adding Elixir to their Next.js chat app for real-time features — why they chose Elixir/Phoenix over Node.js for WebSockets, the architecture trade-offs, and how they run both stacks in production. A candid account of a polyglot architecture decision.

ChatRWKV: ChatGPT Powered by an RNN

ChatRWKV is a ChatGPT-like chatbot built on the RWKV architecture — a language model that achieves transformer-level performance using an RNN design, enabling constant memory inference regardless of sequence length. A significant architectural alternative to attention-based transformers.

TorToiSe TTS: Architectural Design Document

The architectural design document for TorToiSe TTS — James Betker's highly capable open-source voice cloning and text-to-speech system. Explains the multi-model pipeline combining autoregressive and diffusion components that made it state-of-the-art in 2022.

Wide Attention Is The Way Forward for Transformers

What Language Model Architecture and Pretraining Objective Work Best for Zero-Shot Generalization?

Wang, Roberts, Scao et al. (BigScience Architecture Group) conduct a large-scale comparison of model architectures (causal decoder, non-causal decoder, encoder-decoder) and pretraining objectives (autoregressive, masked LM) for zero-shot generalization. The key finding: causal decoders + autoregressive LM win at zero-shot; non-causal decoders + masked LM + multitask fine-tuning win overall.

What Language Model to Train if You Have One Million GPU Hours?

An ablation study by the BigScience group comparing architectural choices and training setups for large multilingual language models targeting 100B+ parameters within a fixed 1M A100 GPU-hour budget. It shows that careful architecture and training setup decisions at the 1.3B scale transfer predictably to larger models, making principled design tractable even at extreme scale.

Unifying Language Learning Paradigms (UL2)

Google Research's UL2 paper proposes Mixture-of-Denoisers (MoD), a unified pre-training objective that combines span corruption, prefix LM, and causal LM into a single framework — and separates architecture choices from pre-training objectives, which were previously conflated. The insight that objective and architecture are orthogonal opened the door to mixing paradigms that were previously treated as distinct camps.

Designing Cloud Data Platforms

Manning textbook on building modern cloud data platforms covering ingestion, storage, processing, and serving layers across major providers. Practical guide for data engineers designing end-to-end pipelines that balance performance, cost, and operational complexity.

Ethereum EVM Illustrated

A visual primer on the Ethereum Virtual Machine internals — accounts, transactions, message calls, gas, the stack/memory/storage architecture, and Go-Ethereum source code. The best single-document mental model for how the EVM actually works at the bytecode level.

High Scalability

High Scalability is a long-running blog covering distributed systems architecture and how companies like Google, Uber, and Meta build systems at scale. Deep technical case studies from production systems — essential reading for engineers designing for scale.

Django for Startup Founders

Alex Krupp's guide to Django architecture for SaaS founders — arguing against the default 'fat models' pattern in favor of service layers, clear separation of concerns, and patterns that scale with a small team. Opinionated and practical.

How to Move Beyond a Monolithic Data Lake to a Distributed Data Mesh

Zhamak Dehghani's foundational essay (hosted on martinfowler.com) proposing data mesh — treating data as a product owned by domain teams rather than centralized in a monolithic data lake. The conceptual origin of the data mesh movement.

The Architecture Behind A One-Person Tech Startup

Anthony Simon's honest account of the infrastructure architecture behind his one-person SaaS business — the actual stack, costs, tradeoffs, and what he'd do differently. Rare for its specificity and honesty about complexity vs. simplicity tradeoffs.

Emerging Architectures for Modern Data Infrastructure

Andreessen Horowitz maps the emerging stack for modern data infrastructure — separating analytics, ML, and operational workloads into distinct architectural patterns. A useful taxonomy of the data tooling landscape circa 2020.

The Complete AWS Lambda Handbook for Beginners

Dashbird's intro to AWS Lambda — the event-driven serverless compute model, execution contexts, triggers, IAM roles, and key gotchas like cold starts and timeout constraints. A solid foundation for serverless architecture thinking.

Data-Oriented Programming

Yehonathan Sharvit's book on Data-Oriented Programming — four principles for separating code from data, using generic structures, embracing immutability, and decoupling schema from representation. A response to the complexity of OOP.

Message Systems in Programming: Callbacks, Events, Pub/Sub, Promises, and Streams

Jesse Warden's survey of async communication patterns in programming — callbacks, events, pub/sub, promises, and streams. A clear taxonomy of a domain where the terminology is often muddled, especially relevant to JavaScript's async evolution in 2014.

Distributed Systems and the End of the API

Prismatic's provocative argument that the traditional synchronous REST API is the wrong primitive for distributed systems — proposing message-passing and event streams as the better foundation. Anticipates the 2014-era shift toward event-driven architectures and the explosion of streaming systems.

VentureBeat Data Architecture Diagram (Screenshot)

Screenshot of a data analysis architecture workflow diagram from VentureBeat, saved alongside the Twitter share of the same image. See the adjacent note on the data analysis architecture diagram.

Data Analysis Architecture Workflow Diagram

VentureBeat/DataBeat data analysis architecture diagram circulated on Twitter in 2013 — a workflow schematic showing the layers of a big data pipeline from collection through analysis to visualization. Snapshot of how practitioners were thinking about data infrastructure.

Designing Machine Learning Frameworks

Mikio Braun's reflection on the design principles behind ML frameworks — the tensions between flexibility and usability, scikit-learn's consistency, and what makes a good ML library API. Written by the author of jBLAS.

Prismatic's Graph at Strange Loop

Prismatic's Strange Loop 2012 talk introducing their Graph library — a Clojure framework for declarative, composable data flow computation where functions are wired together as a directed dependency graph. An early influence on the functional reactive programming ideas that would shape ClojureScript and later frontend architectures.

SQL vs. NoSQL

Linux Journal's SQL vs. NoSQL comparison from the height of the NoSQL hype cycle — a useful grounding piece that distinguished the different NoSQL categories (document, key-value, column-family, graph) and when each made sense. The pendulum has since swung back toward SQL for most use cases.

The Lowline: New York's Underground Park of the Future

The Lowline project proposed converting an abandoned 1908 trolley terminal on Manhattan's Lower East Side into an underground park using fiber-optic sunlight delivery technology. An inverse of the High Line — bringing parkland below street level into existing dead infrastructure.

How Prismatic Deals with Data Storage and Aggregation

Prismatic's 2012 engineering post on their data architecture for real-time news personalization — using Storm for stream processing, Cassandra for distributed storage, and a Clojure-based interest graph. An early look at the stream-processing stack that became standard infrastructure.