Skip to content

GLOSSARY

Terms we actually use.

When a post says 'RAG' or 'invariant test', here's what we actually mean — not Wikipedia, but the project-level explanation.

GLOSSARY

[1/2]

A
AI AgentAI systems
An LLM-driven piece of software that can call tools (APIs, search, DB queries), keeps memory, and solves multi-step tasks. In production every iteration runs guardrails + evals.
ASO (App Store Optimization)Mobile apps
Optimising the app's store listing (title, keywords, screenshots, description, categories). As important as web SEO — just in the App Store / Play Store context.
C
Core Web VitalsWeb & web app
The three metrics (LCP, INP, CLS) Google uses as a ranking factor. We optimise to them on every release.
CSAT (Customer Satisfaction)
A satisfaction score usually on a 1–5 or 1–10 scale. When we ship AI-backed support we measure that CSAT actually climbs after deployment.
D
DevSecOpsCybersecurity
Security as a continuously-running CI step (SAST, DAST, SCA, IaC scan), not an annual project. Runs against every push; every sprint closes at least one security bug.
E
EmbeddingAI systems
A vector representation of text (e.g. 1536 floats). If two embeddings are close, the meanings are close. In RAG we use this to pick relevant chunks.
Eval (LLM evaluation)AI systems
An automated test suite that runs ~50–200 'golden' questions against the model before every release and checks that quality metrics (accuracy, factuality, latency) clear the threshold.
EVM (Ethereum Virtual Machine)Blockchain
The shared execution environment across Ethereum and compatible chains (Arbitrum, Base, Polygon, Optimism). Writing a contract here means it runs on all of them.
Edge renderingWeb & web app
Code runs at the CDN edge closest to the user (Cloudflare Workers, Vercel Edge). Dynamic responses with ~10–50 ms TTFB.
F
Fine-tuningAI systems
Further training an existing LLM on your own data (e.g. LoRA). Rarely the first move — RAG + prompt engineering usually gives more for less cost.
Fuzz testBlockchain
Running millions of pseudo-random inputs through a function to find a combination that breaks it. Echidna or Foundry fuzz campaigns.
FoundryBlockchain
A Rust-based Solidity toolchain (`forge`, `cast`, `anvil`). Faster with better invariant-test support than Hardhat. We run it on every DField audit project.
G
GuardrailAI systems
An input- or output-layer that filters the model's prompt/response (PII scrubbers, prompt-injection detectors, JSON-schema validation, topic blocks). Not before/after the model — around it.
H
Hydration (React)Web & web app
Attaching client-side JS logic to already-rendered HTML. Can be expensive; we reduce it via RSC by hydrating only interactive islands.
I
Invariant testBlockchain
A test that verifies a universal rule ('sum of user balances always equals total supply'). Feeds the contract random transaction sequences; the rule must hold in every state.
ISO 27001Cybersecurity
International standard for Information Security Management Systems (ISMS). Often preferred in Europe instead of or alongside SOC 2. 3-year certification cycle.
ISR (Incremental Static Regeneration)Web & web app
SSG + timed regeneration: the HTML is static but regenerates on a schedule. Ideal for blog content — freshness with CDN speed.
L
LLM (Large Language Model)AI systems
A neural model with billions of parameters (GPT-4, Claude, Mistral) that generates text. In production we never use one bare — always wrapped in retrieval and guardrails.
L2 (Layer 2)Blockchain
A secondary network (Arbitrum, Base, Optimism, zkSync) settling on Ethereum (L1) but much faster and cheaper. Rollups bring 10–100× lower gas.
LCP (Largest Contentful Paint)Web & web app
Time until the largest visible element paints. Google Core Web Vitals passes under 2.5s — we usually land marketing pages under 1s.
M
MEV (Maximal Extractable Value)Blockchain
The value a block-producer can extract by inserting, reordering, or excluding transactions. We factor MEV-resistance into every contract we design.
MFA (Multi-factor auth)Cybersecurity
Two or more factors (TOTP, WebAuthn, biometric) beyond a password. Table-stakes in SaaS today — enterprise procurement disqualifies you without it.
Multi-tenant SaaSWeb & web app
One deployment, many customers (tenants). Data isolation is either row-level (Postgres RLS) or schema-per-tenant. RLS scales to 10k+ tenants; above that, schema.
N
NIS2 directiveCybersecurity
EU cybersecurity directive (in force 2024-10-17) imposing concrete obligations on 'important' and 'essential' organisations — 24-hour incident reporting, supply-chain risk, MFA.
O
OWASP Top 10Cybersecurity
The list of the 10 most critical security risks for web apps, updated yearly. Every DField web pentest starts with this list.
Offline-firstMobile apps
A design philosophy: the app's core functions work without a network and sync when online. Conflict resolution is mandatory, not optional.
P
Penetration test (pentest)Cybersecurity
Manual + tooled attack simulation that reveals what an attacker could achieve. We deliver findings as PRs in your repo, not an 80-page PDF.
R
RAG (Retrieval-Augmented Generation)AI systems
An AI architecture where the model retrieves relevant documents from your own data before answering, and only reasons over that context. Kills ~80% of hallucinations.
RSC (React Server Components)Web & web app
React components that run exclusively on the server and never ship to the browser. Result: less client-side JS and faster hydration.
RLS (Row-Level Security)Web & web app
A Postgres feature that filters rows based on the query context. For multi-tenant SaaS: every SELECT automatically gets a tenant filter appended.
React NativeMobile apps
Meta's framework for building native iOS + Android apps from React-like code, one codebase. Mature since 2020, though ~15–25% is still platform-specific code.
S
Smart contractBlockchain
Code running on a blockchain (Solidity, Rust) that auto-executes an agreement. A single bug can cost nine figures — why we audit, fuzz, and threat-model before deploy.
SolidityBlockchain
The primary language for EVM-compatible blockchains (Ethereum, Arbitrum, Base). Pascal-like syntax, low-level memory handling, TS-compatible toolchain (Foundry, Hardhat).
SOC 2Cybersecurity
A US audit framework for confidentiality, integrity, availability, and privacy controls. For SaaS, the Type II audit (6–12 months of observation) is the standard enterprise baseline.
SSR (Server-Side Rendering)Web & web app
HTML rendered by the server per-request, fresh for every user. Ideal for dynamic content (dashboards), but slower than SSG.
SSG (Static Site Generation)Web & web app
Pages are produced at build time as HTML and served from a CDN. Near-zero TTFB. DField's own site runs this way across 111+ pages.
T
Threat modelCybersecurity
A structured exercise that walks the system's actors, attack surface, risks, and controls. Day one of every DField project — before any code.
TTFB (Time To First Byte)Web & web app
The time until the server's first byte reaches the browser. With edge + CDN we can get this under 50 ms globally.
V
Vector databaseAI systems
A database specialised for fast approximate-nearest-neighbour search over embedding vectors (pgvector, Qdrant, Weaviate). The engineering base of RAG retrieval.
Z
Zero-trustCybersecurity
A security model where every request is suspect (even from inside the network), and every action needs a separate authorisation check. The opposite of the classic perimeter model.
GET STARTED

Let's talk about your project.

Send an email or book a 30-minute call. We work in English and Hungarian.

dezso@dfieldsolutions.com