Do you need a degree or advanced math to become an AI Engineer?

No, you do not need a degree or advanced math to become an AI Engineer, because the job is about building products on top of existing models with a coding agent, and the math it uses is a small set that anyone can learn in a few weeks.

In this blog, we will learn about whether we need a degree or advanced math to become an AI Engineer. We will also see what an AI Engineer actually does day to day, the small set of math the job uses, the math the job does not use, when a degree helps and when a portfolio replaces it, how a researcher role differs from an AI Engineer role in math needs, and how to fill the math gap only as far as the job needs.

We will cover the following:

• What an AI Engineer actually does day to day
• The math an AI Engineer actually uses
• The math an AI Engineer does not need
• When a degree helps
• When a portfolio replaces the degree
• Researcher vs AI Engineer in math needs
• How to fill the math gap only as far as the job needs

I am Amit Shekhar, Founder @ Outcome School, I have taught and mentored many developers, and their efforts landed them high-paying tech jobs, helped many tech companies in solving their unique problems, and created many open-source libraries being used by top companies. I am passionate about sharing knowledge through open-source, blogs, and videos.

I teach AI and Machine Learning at Outcome School.

Let's get started.

What an AI Engineer actually does day to day

An AI Engineer builds products on top of existing models. The model is already trained. We use it through an API and build everything around it: the prompts, the retrieval pipeline (RAG), the agents, the evaluation, and the deployment.

Let's say we are building a support assistant for an online store. A typical day looks like below:

• We describe the task to a coding agent, such as Claude Code, and direct it to write the connector that pulls the store's help articles into a vector database.
• We review the code the coding agent produced, run it, and verify that the right articles come back for a sample question.
• We run the evaluation set of 200 real customer questions and compare the new answers with the old answers.

Here, not a single step needed a formula. The skill is to describe the task clearly, review what the coding agent built, test it, and measure it.

Now, let's see the small set of math that the job uses.

The math an AI Engineer actually uses

There are three things, and each one is a reading skill.

First, reading a probability score.

A classifier that reads a customer message returns a score between 0 and 1, like below:

{ "label": "refund", "score": 0.87 }

Here, 0.87 means the model is confident, and 0.52 means it is barely sure. We use this number to decide a threshold. Suppose we set the threshold to 0.80. Then, every message above 0.80 goes to the refund team, and every message below it goes to a human. We do not need to know how the model computed the score.

Second, what an embedding and a distance between embeddings mean.

An embedding is a list of numbers that represents the meaning of a piece of text. Let's say we have three sentences:

• "How do I return a product?"
• "What is the refund process?"
• "Do you ship to Canada?"

The first two sentences are about the same thing, so their embeddings are close. The third sentence is about something different, so its embedding is far away. The distance is measured with a formula called cosine similarity, which returns a number near 1 for similar texts and near 0 for unrelated texts.

This is the idea behind RAG: we embed the question and pick the help articles whose embeddings are closest to it. We need to understand that close means similar. We do not need to compute cosine similarity by hand, because the vector database does it for us.

Third, basic statistics for evaluation.

When we change a prompt, we must know whether the product got better or worse. Suppose our evaluation set has 200 questions. The old prompt answered 160 correctly, which is 80 percent. The new prompt answered 172 correctly, which is 86 percent. The new prompt is better.

A quick note for you

No matter which tech domain you work in, get familiar with these topics:

• LLM
• RAG
• MCP
• Agent
• Fine-tuning
• Quantization

We put it all together in one video:

AI Engineering Explained: LLM, RAG, MCP, Agent, Fine-Tuning, and Quantization

No need to stop reading - bookmark it and watch later when you get time. Future you will thank you.

Now, let's get back to the topic.

The math an AI Engineer does not need

We do not need to derive backpropagation. Backpropagation is the method used to train a model. The parameters get updated based on how wrong each answer was, and the math behind it is calculus with many variables. An AI Engineer never runs this by hand, because the model is already trained before we touch it. Even when we fine-tune, the training library does the backpropagation and we only choose the data and the settings.

We do not need research-level linear algebra. Inside a model, attention is a set of matrix multiplications. An AI Engineer never writes those multiplications. We need to know only that the model reads every token against every other token, and that this is why a long prompt costs more. Proofs, optimization theory, and information theory are needed to invent a new model architecture, not to build a product on top of one.

When a degree helps

A degree helps in three situations.

First, when we are applying with no work history. A degree in computer science tells the company that a fresher can finish a long, structured piece of work.

Second, when the company has a hard filter. Some large companies require a degree before the application reaches a human. This is a policy of the company, not a requirement of the work.

Third, when the degree taught the fundamentals. Data structures, how a database works, and how to reason about a system are taught in a good computer science program. An AI Engineer uses all of these daily.

Want to become an AI Engineer?

These are the skills this role needs, and we teach them in our AI and Machine Learning Program at Outcome School:

• RAG
• Vector Databases
• AI Agent
• Tool use in Agents
• Evaluation of LLMs and Agents
• Model Deployment and Serving

We take you from the fundamentals to building real AI systems, step by step.

When a portfolio replaces the degree

For anyone with some work history, and for companies without a hard filter, a portfolio does the same work as a degree. A degree says that the person studied. A portfolio says that the person built it, with proof.

A portfolio for this role is built with a coding agent, and each project must show one of the three skills the job uses:

• A RAG assistant over real documents. We direct the coding agent to build the ingestion, the embedding step, and the retrieval, and we show the accuracy before and after a change. This proves that we understand embeddings and distance.
• A classifier with a threshold. We direct the coding agent to build a tool that routes messages using a probability score, and we show how we picked the threshold. This proves that we can read a score.
• An agent with an evaluation report. We direct the coding agent to build an agent that calls two or three tools, and we publish the pass rate on a fixed set of tasks. This proves that we can measure a system.

Each README must state what was built, what was measured, and what the number was before and after.

Stay updated: Subscribe to our newsletter to get our latest AI and Machine Learning blogs straight to your inbox.

Researcher vs AI Engineer in math needs

An AI Researcher invents new models and new training methods. This person derives new loss functions and proves why a method works. Advanced math is the job, and a master's degree or a PhD is usually expected.

An AI Engineer takes the trained model that comes out of that research and builds a product with it. For this role, advanced math is not the job. The job is system design, agentic coding, evaluation, and deployment.

Let me tabulate the differences between the two roles for your better understanding.

For the researcher role, we need the degree and the math. For the AI Engineer role, we need the portfolio and the three reading skills.

How to fill the math gap only as far as the job needs

We fill the gap in the order the job uses it, and we stop when the job stops needing it.

Step 1: Learn to read a probability score. Spend one week building a small classifier with a coding agent, print the score for 50 messages, and move the threshold up and down.

Step 2: Learn what an embedding is. Spend one week embedding 20 sentences with the coding agent, computing the cosine similarity between pairs, and checking that similar sentences score near 1.

Step 3: Learn evaluation statistics. Spend one week directing the coding agent to build an evaluation set of 100 questions for a search over those sentences, computing accuracy, and comparing two versions of the retrieval.

After these three weeks, we have every piece of math that the role uses and three projects to show for it.

If one day we move toward the researcher role, we learn the advanced math then, because at that point it becomes the job.

Now we must have understood that an AI Engineer does not need a degree or advanced math, which small set of math the job uses, and how to fill the gap only as far as the job needs.

Frequently Asked Questions

Can a fresher without a computer science degree become an AI Engineer?

Yes, but this is the case where a degree helps most. When we apply with no work history, a degree in computer science tells the company that a fresher can finish a long, structured piece of work. Some large companies also require a degree before the application reaches a human, but this is a policy of the company, not a requirement of the work. For companies without such a filter, a portfolio does the same work as a degree, because it shows that we built something, with proof.

Do we need to compute cosine similarity by hand as an AI Engineer?

No. We do not need to compute cosine similarity by hand, because the vector database does it for us. We only need to understand that close embeddings mean similar text. Cosine similarity returns a number near 1 for similar texts and near 0 for unrelated texts.

Do we need to know backpropagation to fine-tune a model?

No. Even when we fine-tune, the training library does the backpropagation for us. We only choose the data and the settings. Backpropagation is the method used to train a model, and its math is calculus with many variables, but an AI Engineer never runs it by hand.

How long does it take to learn the math an AI Engineer uses?

It takes about three weeks. We spend one week on reading a probability score, one week on embeddings and cosine similarity, and one week on evaluation statistics. After these three weeks, we have every piece of math that the role uses and three projects to show for it.

That's it for now.

Thanks

Amit Shekhar
Founder @ Outcome School

You can connect with me on:

• X
• LinkedIn
• GitHub

Follow Outcome School on:

• X
• YouTube
• LinkedIn
• GitHub

Read all of our high-quality blogs here.