I list the core pieces of knowledge and basic skills that you need to start being a good engineer. They seem numerous but without them you will be like those who did not even know that a wheel exists somewhere across the ocean. You do not need to become an expert in each topic, just getting the figers dirty is enough in some of them. The list is divided into Technologies, Math, Languages, Organization, Trends, Beyond Engineering.
But before you start reading, I must answer the natural question of whether it is necessary to be an engineer in the nearing age of AGI or specialized AI. The secret sauce of being an engineer is not in creating new things but in creating new things efficiently. It is possible to train a model to choose the suitable technology and to code efficiently. And it is possible that at some point such models will not need human involvement. So, maybe long term there won't be a need in human mathematicians and coders. Although another part of me says that we will always want to keep an eye on the machines and to do so, we need to be good mathematicians, scientists and engineers with the help of machines in areas where they can help us with their speed and access to information.
A good engineer should know how things work inside. If you write code and do not know how it is compiled and executed, you will not be able to design efficient products. To rely on the designers of languages, libraries, containers and operating systems, clouds to do the best job, you must know how these products are designed and structured in order to choose a correct produst and use it efficiently. For example, every language has concurrency but their design goals and correspondingly implementations are so different. We'll get to this below.
I prefer to learn from bottom up and therefore I'll start with hardware.
Keep in mind that in order to know something, you have to do it. Therefore most sections describe not only what you should know but also what you should try.
At the end of the day, whatever you write runs on hardware and if you want to be methodical in learning engineering, start with hardware.
Take a simple controller, such as Arduino, with 1 sensor and 1 output device and write a loop that does something simple.
No need to dive deeper unless you want to go into firmware engineering which actually is a great field due to the rapid development of robotics, automation, AI.
To learn Aeduino I recommend buying a kit like Arduino Student Kit, using the online IDE and following the examples.
Learn C which is the foundation for today's mostly used languages. If you take Arduino route described above, you'll learn basic C.
Learn what types of computer languages there are. What considerations go into the design of languages. Wikipedia has a nice historic overview and structural description here
There are several approaches to parsing and compiling programming languages. Learn about them. Design a simple language that has int, float, conditions, loops and functions. Then implement it. There are online articles that explain how to build your own lexer, parser, interpreter or compiler both from scratch and using existing products. Building anything from scratch in C is better for understanding how things work.
This series of articles shows how to implement a new language in C++ step by step. Also check ANTLR which has become the most popular and easiest way of constructing languages.
Learn how Linux manages
- multiple processes and threads. Process and thread priority. Core affinity. How Linux handles a process & thread when it's doing IO. What hyperthreading is.
- memory
- IO (devices and data)
- file systems
Learn basic shell commands that show information about and manage processors, memory, file systems and files, network, devices, processes and threads.
Extra points for looking at the source of these shell commands, and of Linux core.
Learn:
- how Linux manages processes and threads as described above in Linux section.
- how to use processes & threads in Rust or in Java, or in any language that uses real OS threads (not Dart, JS, Python, Go, Erlang)
- how processes exchange data
- how to decide when to use processes and when threads (hint: % of IO, size of exchanged data)
- concept of threads pools and why they are needed (hint: % of IO + number of cores)
- difference between parallelism and concurrency
- how concurrency is implemented in JS and Python. Important: not how to use it but
- what happens to your Python code when you run it in multiple threads (hint: GIL and why this design was chosen originally)
- what happens to your JS code when you create a callback (hint: event loop and why this design was chosen by creators of V8)
- concept of "green threads" and how concurrnecy is implemented in Go. Important: not how to use goroutines but how Go executes them, queues them, switches among them.
Exercise: calculate the total size of all files on a disk without recursion (hint: use a queue) and using concurrency in JS, Python, Rust/Java and Go/Erlang (they implement concurrency differently). In each of the 4 languages: what was faster: threads or processes and why.
Learn IP4 and TCP protocols. There are many more commonly used protocols but these 2 are the base for reliable exchange of data over the internet.
Exercise: write a packet sniffer in C
Learn TLS. Security goes beyond encrypting traffic but learning about TLS covers multiple topics in security and serves as a good base.
Exercise: implement an HTTPS client in C
In addition, read about cryptographic hashing. It is commonly used and not only for protecting data.
Docker is used to create an environment for coding and running stuff. For example, on your Mac or Windows machine you can code something in a Linux environment and then deploy it onto a "cloud" (AWS, Azure, GCP, Oracle etc.). This creates consistency among engineers and environments (development, testing, staging, production). Nowadays it's a must.
Learn about the most commonly used Linux filesystems: their design, their purpose, when to use which.
Learn about the types databases:
- relational (MariaDB)
- document (MongoDB)
- columnar (Cassandra)
- data structures (Redis)
- key/value (Memcached)
- query engines (PrestoDB)
- graph (Neo4J)
- search (ElasticSearch)
- vector databases which are relevant for Machine Learning. An explanation. List of open source vactor databases not in a particular order:
- blockchain (see below)
Answer these questions for each:
- how they store and retrieve
- how they scale with
- concurrent inserts, updates, deletes, retrievals and why
- size of data for inserts, updates, deletes, retrievals and why
It is good to get a hands-on experience with each of them: install, insert data, retrieve data;
It is good to do the same for their cloud-based serverless analogs. AWS is the most widely used cloud.
Global DB ranking. You can also rank them by type. They classify Radis and PrestoDB differently than I do, so follow my list.
At your current and next job you will not use all these database types but knowing the types and their strong/weak sides is good to be able to choose the best when you need.
Learn the difference between vertical and horizontal scaling.
Vertical scaling is done by
- hardware that better addresses the performance bottlenecks
- choice of software and its configuration
- code optimimizations if you are writing it
Learn how the database technologies listed above implement horizontal scaling, how they decide on which node to store which part of data etc. They are quite different. Even through we code on Clouds, knowing how clouds scale is useful because you may work on one of them, not all technologies are on your cloud, it is useful to know how your technology of choice scales for your efficient use of it.
Create an AWS account and do everything above on AWS as if your local computer is just a terminal.
AWS because it is the default. If you need to work on another cloud, it will be similar.
Now when we are on AWS, learn Kubernetes: what it is, why it was created, its components, how to run it locally. Create a basic echo server and deploy it locally into 2 pods with a load balancer and try to use it.
Learn Helm Charts for K8s and do the same ^ using Helm.
Then deploy it onto AWS EKS.
GPU is a part of most phones and laptops. It is good to know how GPU architecture is different from CPU. Due to GPU availability and GPU's ability to process matrices, whether to use GPU for parts of code should be a consideration in your software design.
Before learning about GPU, it is useful to learn about matrices and vectors.
Also read about General-Purpose computing on Graphics Processing Units, OpenCL and take a look at this Python example
You need to know how to build a Web UI in a few cases:
- this is your specialization
- your current/next role requires this skill, often as a part of a "full stack" role
Otherwise, knowing how to build a Web UI is frequently needed to visualize what you work on, especially if the data is dynamic.
Nowadays most Web Applications are built using frameworks such as ReactJS but I am not sure that one can build anything using a framework without knowing the following building blocks:
Learn how negative and floating point numbers are represented in commonly used processors.
CS is the mathematical principles for designing efficient technologies.
To get the flavor of what I mean study how the core data structures are implemented in C: array, linked list, hash map, trees, graphs, probabilistic data structures.
Learn how to calculate the time to add an element and to find an element using big O notation.
Learn at least 3 types of sorting algorithms for each data structure, if applicable.
Any good design must use math. Otherwise it has unexpected consequences.
OO Design is a must for designing complex systems. Use C++ when you learn this. There are languages with better OO design but C++ is a good first step in this direction after C.
Learn "design patterns". There are also non OO design patterns.
At its essence a blockchain is a distributed decentralized transaction log. This is why I listed it in Storage section. If you need to store all previous states of a record in an unmodifiable, verifiable, resilient and scaleable way, you can deploy the code of an existing blockchain, you'll just need to find a blockchain with features that you need.
Statistics helps us analyze data. For example, data coming from logs. That's why it is useful to know basic statistics definitions and tools.
Terms:
- histogram
- types of mean/average
- standard deviation
- normal distribution
- covariance
- r-squared
- correlation
- enthropy
- normality test
Tools
The commonly used languages should have these statistical functions ether as a part of the standard library or as an open source library.
These languages are commonly used by data engineers, analysts, scientists and mathematicians:
- Python. To work with tabular data use Pandas/Dask. Also, they have statistical functions. If you need more statistics, use statsmodels.
- R
- Julia
Jupyter: a tool for analytsing data.
Like statistics, ML is a useful tool that, if you are familiar with, may increase your options of data analysis. It is also one of the directions in which technology is currently developing. So, not being familiar with ML is a disability.
You can learn ML as
- an engineer: most ML projects share the same steps from gathering data all the way to monitoring a model in production. AKA ML Workflow
- as a data scientist: know how each step of the ML Workflow works. For example, how to identify if data is suitable to be used for ML.
- as an ML user: use a library that does everything for you.
ML deserves a separate page. TBD +TPU/NPU,LPU
In addition, understand the difference between AI and ML.
I recommend learning these languages no matter what:
- Assembley: if you haven't tried Assembley, you do not know how a computer works.
- C: if you haven't coded in C, you do not know how software actually works.
- C++: low level OO language used for firmware and for speed
These languages are good for their respective areas:
- JavaScript for APIs and web applications
- Dart for mobile apps
- Python for data and for APIs/backend when speed is of lower importance
- Julia for data if you also need speed and concurrency. It is also good for APIs with spikes due to its coroutines.
Here is more info.
It is good to learn the inner structure of the leading AI tech. Otherwise you do not really know what it is. It may also provide you with interesting job opportunities.
Use an AI for coding and research. At the moment Gemini generates the best code.
Learn what is Unit Testing versus Integration Testing. What their best practices are. Learn how to set them up in the languages and for the technologies that I listed above.
The worst engineers that I see are those whose goal is to accomplish goals. This sounds counterintuitive. Who would not like an employee who achieves goals. This is true and it works in many cases. But achieving a goal is not the same as producing scalable code, i.e. modular, readable, easy to test, deploy and monitor, scalable in speed and throughput.
The best engineers that I see are those who do engineering in addition to a job because doing it inspires them.
The best engineers want to see the bigger picture: what is the goal of my task, what are the effects of my code.
The best engineers look for improvements in the requirements are argue with their bosses.
Good engineers love to learn, improve, create, see the positive effect of their work.
Better engineers want their teammates and team to succeed. You can work for your boss (bad engineer from above), work for yourself, work for the love of the art of engineering, or you can work for the success of your teammates, your team, your company, mankind. Working for the succees of others does not exclude the other levels but puts you on a higher level as an engineer and human. It takes time to grow to this level but at least keep this in mind.
You should master 2 types of workflow: task and team.
Working on a task means going through steps such as design, implementation, testing, deployment, monitoring (there are more steps and sub-steps). You will naturally progress through some of them starting with your first Assembley project. Some of them come only when you work on a real product for real users.
Another type of workflow is how a team organizes its works. It is very specific for a team and it comes with experience but from your first Assembly project you should follow the Agile Principles. Learn about them, apply them, review how efficient you were, think of improvements, repeat.
Know the current tech trends. At the moment these seem worthy:
- 5G: learn how different it is from 4G, what new opportunities it opens
- Quantum Computing: what it is on the physical level, how it can be used
- Metaverses: keep an eye on them. At some point when companies start to cooperate on the network protocol and data format, we'll go 3D.
- AI
Engineering skills can be applied to many other situations in life because we should be good at learning situations and paving a path to a desired outcome. This may sound as a no-brainer but many people have mental blocks. Also, the ability to come up with ideas, the desire to do something useful, the social skills are applicable anywhere.
Be an engineer: free, all powerful and positive for others.