Glossary D - I

Reusable solutions to commonly occurring problems in software design, originally documented by the Gang of Four across creational, structural, and behavioral categories.

Autonomous AI agent for software development by Cognition AI. Costs $500/month and is used by Goldman Sachs on $2M+ projects.

Set of practices combining software development and IT operations to shorten the development lifecycle and deliver changes more frequently and reliably.

Set of strategies and procedures for restoring critical systems and data after a catastrophic event, minimizing downtime and information loss.

Observability technique that tracks the complete journey of a request across multiple services in distributed architectures to diagnose performance issues.

Platform for building, shipping, and running applications in isolated containers that package code and dependencies consistently.

Software design approach that centers architecture around the business domain and its logic, using bounded contexts and ubiquitous language to manage complexity.

Set of four key metrics measuring engineering team software delivery performance: deployment frequency, lead time, MTTR, and change failure rate.

Alignment method that optimizes language models directly from human preferences without requiring a separate reward model, simplifying the RLHF process.

DRY (Don't Repeat Yourself) is a software development principle establishing that each piece of knowledge should have a single authoritative representation in a system.

AWS fully managed NoSQL database designed for high-performance applications with single-digit millisecond latency and automatic scaling.

Numerical vector representation of data in a continuous space where similar items are positioned closer together.

Large body of work in agile methodologies grouping multiple related user stories under a common business objective, spanning several sprints.

Storage pattern that persists entity state as an immutable sequence of events rather than storing only the current state.

Software design pattern where program flow is determined by events such as user actions, service messages, or state changes.

Configuration mechanism that enables or disables features at runtime without deploying new code, enabling gradual rollouts, A/B testing, and safe deployments.

Process of further training a pre-trained model on domain-specific data to improve performance on particular tasks.

Cloud financial management discipline that combines finance, technology, and business to optimize cloud spending.

LLM capability to generate structured function calls that can be executed by external systems.

CI/CD platform built into GitHub that automates build, test, and deployment workflows directly from the repository.

AI coding assistant by GitHub/Microsoft that provides inline code suggestions, chat, and agent capabilities within IDEs.

Operational framework that uses Git as the single source of truth for declarative infrastructure and application configuration.

Family of LLMs by OpenAI that pioneered the transformer architecture for text generation.

Query language for APIs that allows clients to request exactly the data they need, reducing over-fetching and under-fetching.

Technique of anchoring LLM outputs to verifiable sources of truth like databases, documents, or APIs to reduce hallucinations.

High-performance remote communication framework developed by Google that uses Protocol Buffers and HTTP/2 for efficient service-to-service communication.

Safety mechanisms that constrain AI model behavior within acceptable boundaries, including input/output validation, content filters, and policy enforcement.

When an LLM generates confident but factually incorrect or fabricated information. A key challenge in production AI systems.

Engineering practice where every time an agent makes a mistake, a solution is designed to ensure the agent never makes that mistake again. Concept popularized by Mitchell Hashimoto in Ghostty development.

Decoupled architecture where the backend manages content and data without imposing a presentation layer, allowing any frontend to consume it.

Architectural pattern that isolates business logic from the outside world through ports and adapters, improving testability and enabling technology changes.

Scalability strategy that adds more machines to distribute workload across multiple servers instead of upgrading a single one.

Property of an operation that produces the same result regardless of how many times it is executed, fundamental for reliability in distributed systems.

Set of techniques for reducing latency, cost, and resource consumption when running AI models in production, including quantization, batching, and caching.

Server, network, and database management through versionable configuration files (Terraform, Pulumi).