Software project terms, in plain English

Each term below has one definition and a link to where it applies. Terms are listed in the order a project meets them.

Hiring and delivery models

Staff augmentation: Adding named engineers to your team, who work under your manager in your process. Compare it with a pod in dedicated team or staff augmentation.
Offshore development centre: A unit of engineers in another country that works in your name and process, while the supplier handles employment, premises and equipment. See the delivery centre service.
Build-operate-transfer: A contract in which a supplier builds a team, runs it for an agreed period, then hands it to you. The transfer date, price and method are written into the first contract. See build-operate-transfer.

Discovery phase: The first, self-contained stage of a project, where architects and analysts study how the work runs today. The process maps and findings belong to you even if you stop there.
Solution blueprint: The written plan that comes out of discovery: the problem, the target design and the phased steps, with points where you can stop. You can hand it to any team to build.
Build versus buy: The choice between commissioning software made for you and licensing a product. See build or buy.
Total cost of ownership: Everything a system costs over its whole life: purchase or build, hosting, upkeep, licences, support and your own staff time.

Enterprise resource planning: One system for a company's orders, stock, production, finance and people, so each team reads the same figures. See custom ERP or SAP.
System of record: The one system whose version of a fact, such as a customer's address, wins when other systems disagree.
Master data: The long-lived reference records every process uses: items, customers, suppliers, price lists and cost centres.
Canonical data model: A single agreed shape for shared records, so each system converts to and from it once instead of to every other system.
Multi-tenancy: One running copy of software that serves many customers, each seeing only its own data.
Data residency: The rule or choice that data stays in a named country or region, including its backups and copies. See compliance.
DPDP Act 2023: The Indian law on personal data in digital form. It covers consent, the stated purpose of collection, how long data is kept, and what rights each person has over their own data. See the Act and the DPDP checklist.

Legacy modernisation: Updating or replacing an ageing business system in steps, so the business keeps running while the change happens.
Strangler pattern: Moving a system one function at a time into new software, so the old system shrinks until it can be switched off.
Parallel run: Running old and new systems side by side on the same work, and comparing their results, before the old one is retired. See Tally to a custom ERP.

Retrieval-augmented generation: A way of using a language model where the system first fetches passages from your own documents, then asks the model to answer from them, citing the source.
Evaluation set: A fixed list of questions with known correct answers, used to score an AI system before and after each change.
AI guardrails: Limits built around an AI system, such as budget caps, lists of allowed actions, review lanes for low-confidence answers and a kill switch.
Service level objective: A measurable target for a service, such as the share of requests answered within a set time, agreed before launch.
Hypercare: The period right after launch when the build team keeps working at full pace on whatever real use exposes, before routine support takes over.

The first call is 45 minutes with an architect, usually the founder. It is free, and we will sign an NDA first if you ask.