You do not need to write code to work in software. You do need to know what the words mean, because they are used constantly and nobody stops to define them.
This lesson gives you one picture and hangs the vocabulary off it. Almost every system you will hear discussed is a variation on that picture.
Learning Objectives
After this lesson, you will be able to:
Sketch the universal architecture: client, API, services, data, infrastructure
Explain what an API is, and what REST, GraphQL and webhooks each solve
Tell relational and non-relational databases apart, and say when each fits
Distinguish monolith from microservices, and say why starting with a monolith is usually right
Read the cloud service models, and decode a job description's technology list
The value of holding one picture is that new vocabulary stops being new. When someone mentions a technology you have not heard of, the useful question is which box it sits in. It is nearly always a different answer to a question one of these boxes already asks.
The frontend is what the user sees. On the web that is HTML, CSS and JavaScript, usually built with a framework such as React, Angular or Vue. On mobile it is Swift for iOS, Kotlin for Android, or a cross-platform tool like React Native or Flutter that targets both.
The backend holds the rules. Common languages include Java, Python, C#, Go and JavaScript running on Node. This is where "when an order is cancelled within thirty minutes, refund automatically" actually lives.
The API is the contract between them. It defines what can be asked for and what comes back, which means the frontend and backend teams can work independently as long as both honour it.
Databases hold the record of what is true. Queues hold work to be done later, so a slow job does not make the user wait.
An API, or application programming interface, is how two pieces of software talk. You will meet four kinds.
REST is the default. Requests go to URLs and data comes back as JSON. Most APIs you meet will be REST, and it is worth knowing that a REST response carries a status code: 200 means it worked, 404 means the thing was not found, 500 means the server broke.
GraphQL lets the client ask for exactly the fields it wants in one request, rather than calling three endpoints and discarding most of what comes back. It solves a real problem and adds real complexity.
gRPC is used for fast communication between services inside a system rather than with the outside world.
Webhooks invert the direction. Instead of you asking repeatedly whether something has happened, the other system calls you when it does. Payment providers use them heavily.
APIs are usually documented with OpenAPI, often still called Swagger, and exercised by hand with a tool like Postman.
The first split is relational against everything else.
Relational databases store data in tables with defined columns and enforce relationships between them. You query them with SQL. PostgreSQL, MySQL, Oracle and SQL Server are the common ones. They are the right default for most systems, and the reason is that they enforce structure and guarantee that a transaction either fully happens or fully does not.
Non-relational, usually called NoSQL, covers several quite different tools grouped by what they are not. Document stores like MongoDB hold flexible records. Key-value stores like Redis are extremely fast and often used as a cache. Wide-column stores like Cassandra handle enormous write volumes. Search engines like Elasticsearch exist to answer text queries quickly.
Separately, a data warehouse such as Snowflake or BigQuery holds data for analysis rather than for running the application, and pipelines move data into it on a schedule.
What Do You Think?
A team is building a banking system that moves money between accounts. Which database choice is the sensible default, and why?
A monolith is one deployable application containing everything. One codebase, one deployment, one thing to run.
Microservices split the system into many small independent services, each owned and deployed by one team, communicating over APIs or events.
Microservices exist to solve an organisational problem more than a technical one. When forty engineers share one deployment, they queue behind each other. Splitting the system lets teams move independently.
The cost is that a function call becomes a network call. Things that could not previously fail now can. Debugging a single user request may mean tracing it across six services, which is why teams that go this way invest heavily in monitoring.
Cloud providers rent computing rather than selling it. The four models differ in how much you manage.
Model
You manage
The provider manages
Examples
Infrastructure as a service
The operating system, your application, your data
The physical servers, network and storage
Virtual machines on AWS or Azure
Platform as a service
Your code and your data
Everything underneath, including scaling
App hosting services
Software as a service
Only your usage of it
The entire application
Email, CRM, ticket tools
Serverless
Individual functions of code
All servers, scaling down to nothing when idle
Hosted function services
The three large providers are AWS, Microsoft Azure and Google Cloud. Which one a company uses is frequently a historical accident, and the concepts transfer between them.
Two words you will hear alongside. A region is a geographic location where your system runs, which matters both for speed and for laws about where data may be stored. Availability zones are separate facilities within a region, so a system can survive one of them failing.
Quick Check1 / 3
A job description lists React, FastAPI, PostgreSQL, Redis and AWS. What does that tell you about the role?
One picture covers almost everything: client, frontend, API, backend services, data stores, infrastructure. New vocabulary is a new answer to one of those boxes.
An API is a contract. REST is the default, GraphQL shapes what the client receives, webhooks reverse who initiates, gRPC is for service-to-service speed.
Relational databases are the sensible default because they enforce structure and guarantee transactions. Non-relational tools each solve a narrower problem well.
Microservices solve a team-coordination problem, not a technical one. Start with a monolith and split under real pressure.
The cloud models differ by how much you manage. Regions and availability zones are about geography, law and surviving failure.
Next: The Career Playbook — how to sound experienced, what certifications actually signal, and the unwritten rules worth pinning up.