Be an ML Engineer, not an API Caller

Be an ML Engineer, not an API Caller

Why aspiring Machine Learning Engineers need strong fundamentals.


12 min read

In the rapidly evolving landscape of artificial intelligence, Large Language Models (LLMs) like OpenAI's ChatGPT and Google's Gemini have taken center stage. These powerful models, accessible through simple API calls, have opened up new possibilities for startups and developers, enabling them to integrate advanced natural language processing capabilities into their applications with ease. However, amidst the excitement and hype surrounding these LLMs, a concerning trend has emerged: aspiring Machine Learning Engineers (MLEs) are jumping on the bandwagon without acquiring a solid foundation in the core concepts of machine learning. In this blog post, we will delve into the reasons why relying solely on LLM APIs is a misguided approach for beginners, explore how companies are capitalizing on the hype, and emphasize the crucial importance of strong fundamental ML knowledge for senior roles.

The Allure of LLM APIs

The rise of LLMs has undeniably revolutionized the field of natural language processing. These models, trained on vast amounts of textual data using state-of-the-art deep learning architectures, have demonstrated remarkable capabilities in generating human-like responses, answering questions, and assisting with a wide range of tasks, from content generation to code completion. The ease of accessing these capabilities through simple API calls has attracted many developers, especially those with backgrounds in web development or other programming domains. The promise of quickly building intelligent applications without the need to delve into the intricacies of machine learning has proven to be a tempting proposition for many.

The allure of LLM APIs is understandable. They provide a seemingly straightforward path to incorporating cutting-edge natural language processing capabilities into applications, without the steep learning curve associated with traditional machine learning approaches. For developers who are already proficient in web technologies and API integration, leveraging LLMs can feel like a natural extension of their skill set. The ability to create chatbots, content generators, and intelligent assistants with just a few lines of code is undeniably appealing.

Moreover, the hype surrounding LLMs has been fueled by the impressive demos and case studies showcased by companies like OpenAI and Google. The media attention and buzz generated by these showcases have further contributed to the perception that LLM APIs are the key to unlocking the potential of machine learning. Developers, eager to ride the wave of excitement and build innovative applications, are easily drawn to the simplicity and power promised by these APIs.

The Pitfalls of Overreliance on APIs

While LLM APIs provide a convenient entry point into the world of ML, relying solely on them can be detrimental to the long-term growth and success of aspiring MLEs. Let's explore the reasons why:

  1. Lack of Understanding: When developers simply call APIs to leverage the capabilities of LLMs, they miss out on the opportunity to understand the underlying principles and algorithms that power these models. They remain oblivious to fundamental concepts like backpropagation, activation functions, gradients, batch normalization, and attention mechanisms, which form the backbone of modern ML systems.

Without a solid grasp of these concepts, people are essentially treating the LLM APIs as black boxes. They input data and receive outputs without comprehending the intricate processes that occur behind the scenes. This lack of understanding can lead to several problems down the line, such as difficulty in troubleshooting issues, inability to optimize performance, and limited ability to adapt the models to specific domains or requirements.

Moreover, relying solely on APIs can create a false sense of expertise. People may believe that they have mastered machine learning simply because they can integrate LLMs into their applications. However, this superficial understanding is far from the deep, foundational knowledge required to truly excel in the field of machine learning.

  1. Limited Customization: LLM APIs offer pre-trained models with fixed architectures and parameters. While these models can be fine-tuned for specific tasks, the ability to customize and optimize them is limited. MLEs who rely solely on APIs may struggle when faced with unique problem domains that require bespoke model architectures or training techniques.

In real-world scenarios, off-the-shelf LLMs may not always provide the best solution for a given task. Different applications and domains often require customized approaches, such as domain-specific preprocessing, data augmentation techniques, or novel model architectures. Without a deep understanding of ML concepts, aspiring MLEs may find it challenging to adapt and customize LLMs to meet the specific needs of their projects.

Furthermore, the performance of LLMs can vary significantly depending on the quality and relevance of the training data. If the pre-trained models are not well-suited for a particular domain or task, simply calling the API may yield suboptimal results. MLEs who lack the knowledge to fine-tune and optimize these models may struggle to achieve the desired performance levels.

  1. Dependency on Third-Party Services: Building applications that heavily depend on external APIs leaves developers vulnerable to changes in pricing, availability, and terms of service. If the API provider decides to modify their offerings, update their pricing structure, or discontinue the service altogether, the entire application may be at risk.

This dependency on third-party services can create significant challenges for developers. If an application is built around a specific LLM API, and that API becomes unavailable or prohibitively expensive, the developer may be forced to undertake a major overhaul of their system. This can be a time-consuming and costly process, especially if the developer lacks the fundamental ML knowledge to quickly adapt to alternative solutions.

Moreover, relying on external APIs can raise concerns about data privacy and security. When sending sensitive data to third-party services for processing, developers must carefully consider the potential risks and ensure that appropriate safeguards are in place. Without a deep understanding of the underlying technologies and best practices in data handling, developers may inadvertently expose their applications and users to security vulnerabilities.

  1. Inability to Troubleshoot: When issues arise with the performance or output of an LLM, developers who lack a deep understanding of ML concepts may find themselves ill-equipped to diagnose and resolve the problems. Without knowledge of the underlying mechanisms, they are left at the mercy of the API provider's support or documentation.

Troubleshooting machine learning models can be a complex and iterative process. It requires a solid understanding of the model architecture, training process, and evaluation metrics. MLEs who rely solely on APIs may struggle to identify the root causes of issues, such as overfitting, underfitting, or data bias. They may not have the tools or knowledge to perform effective error analysis, interpret model predictions, or implement appropriate remediation strategies.

In contrast, MLEs with strong fundamental knowledge can approach troubleshooting in a more systematic and effective manner. They can analyze the model's behavior, examine intermediate outputs, and make informed decisions about adjusting hyperparameters, modifying architectures, or collecting additional training data. This ability to diagnose and resolve issues is crucial for building robust and reliable ML systems.

The Role of Companies in the Hype

It's important to recognize that companies like OpenAI and Google have a vested interest in promoting their LLM APIs. By making these powerful models accessible to the masses, they aim to expand their user base, gather valuable data, and ultimately monetize their offerings. While these companies have undoubtedly made significant contributions to the field of ML, their marketing strategies often gloss over the importance of foundational knowledge and the limitations of relying solely on APIs.

These companies often present their LLM APIs as a panacea for all natural language processing tasks, promoting the ease of integration and the impressive results that can be achieved with minimal effort. They showcase compelling demos and case studies that highlight the capabilities of their models, creating a sense of excitement and potential among developers.

However, what is often understated in these promotions is the importance of having a strong foundation in machine learning concepts. The marketing materials may not emphasize the limitations of relying solely on APIs or the potential pitfalls that can arise when developers lack a deep understanding of the underlying technologies.

Moreover, companies may not always provide comprehensive guidance on best practices for using their APIs, such as data preprocessing, model evaluation, or error handling. Without this guidance, people may struggle to use the APIs effectively and may not be aware of the potential risks and challenges associated with building applications around these models.

It's crucial for people to approach the hype surrounding LLM APIs with a critical eye. While these APIs can indeed be powerful tools, they should be seen as a complement to, rather than a replacement for, strong fundamental ML knowledge. Companies have a responsibility to promote the importance of foundational learning alongside the convenience of their APIs, ensuring that aspiring MLEs have a well-rounded understanding of machine learning concepts.

The Necessity of Strong ML Fundamentals

As aspiring MLEs progress in their careers and seek senior roles, the importance of strong fundamental ML knowledge becomes increasingly evident. Let's explore the reasons why:

  1. Algorithm Selection and Design: Senior MLEs are often responsible for selecting the most appropriate algorithms and designing custom model architectures for specific tasks. This requires a deep understanding of the strengths and weaknesses of different ML techniques, as well as the ability to evaluate their suitability for a given problem domain.

Without a solid grasp of ML concepts, making informed decisions about algorithm selection and model design becomes challenging. Senior MLEs need to be able to reason about the trade-offs between different approaches, such as the choice between deep learning and traditional machine learning algorithms, the selection of appropriate neural network architectures, or the application of domain-specific techniques like transfer learning or few-shot learning.

Moreover, designing effective model architectures requires an understanding of the underlying mathematics and principles of machine learning. Senior MLEs should be familiar with concepts like loss functions, optimization algorithms, and regularization techniques, as well as the intricacies of different neural network layers and activation functions. This knowledge allows them to make informed decisions about model design, ensuring that the chosen architecture is well-suited for the task at hand and can be trained efficiently.

  1. Model Optimization: Optimizing models for performance, efficiency, and scalability is a critical skill for senior MLEs. This involves a deep understanding of techniques like hyperparameter tuning, regularization, and model compression.

Hyperparameter tuning is the process of selecting the optimal values for the various parameters that control the behavior of an ML model. This can include settings like learning rate, batch size, number of hidden layers, and activation functions. Effective hyperparameter tuning requires an understanding of how these parameters impact model performance and the ability to systematically explore the hyperparameter space to find the best combination of values.

Regularization techniques, such as L1 and L2 regularization, dropout, and early stopping, are used to prevent overfitting and improve the generalization ability of ML models. Senior MLEs need to understand the principles behind these techniques and how to apply them effectively in different scenarios. They should be able to evaluate the trade-offs between model complexity and generalization performance and make informed decisions about the appropriate level of regularization to use.

Model compression techniques, such as pruning, quantization, and knowledge distillation, are used to reduce the size and computational requirements of ML models while preserving their performance. Senior MLEs should be familiar with these techniques and know when and how to apply them to optimize models for deployment in resource-constrained environments, such as mobile devices or embedded systems.

  1. Research and Innovation: Contributing to the advancement of the field of machine learning requires a solid grasp of the underlying principles. Senior MLEs with a strong theoretical foundation are better equipped to conduct research, propose novel approaches, and drive innovation in the domain of machine learning.

Research in machine learning often involves exploring new architectures, training techniques, or optimization algorithms. To make meaningful contributions in this area, senior MLEs need to have a deep understanding of the current state-of-the-art and the ability to identify gaps or limitations in existing approaches. They should be able to formulate research questions, design experiments, and evaluate the effectiveness of proposed solutions using rigorous statistical methods.

Innovation in machine learning also requires creativity and the ability to think outside the box. Senior MLEs with a strong fundamental knowledge are better positioned to come up with novel ideas and approaches that push the boundaries of what is possible. They can draw upon their understanding of the underlying principles to propose new architectures, loss functions, or training strategies that address specific challenges or improve upon existing techniques.

Moreover, keeping up with the rapidly evolving field of machine learning requires continuous learning and the ability to critically analyze new research papers and techniques. Senior MLEs with a solid theoretical foundation are better equipped to understand and evaluate the contributions of new research, assess the potential impact on their own work, and incorporate relevant ideas into their projects.

  1. Collaboration and Communication: Senior MLEs often collaborate with cross-functional teams, including data scientists, software engineers, and domain experts. Having a shared understanding of ML concepts facilitates effective communication, knowledge sharing, and problem-solving.

In these collaborative settings, senior MLEs need to be able to explain complex machine learning concepts to non-technical stakeholders, such as business leaders or product managers. They should be able to articulate the strengths and limitations of different approaches, discuss the trade-offs involved in model selection and optimization, and provide insights into the potential impact of machine learning solutions on business objectives.

Moreover, collaboration with data scientists and software engineers requires a common language and understanding of machine learning principles. Senior MLEs should be able to engage in technical discussions, provide guidance on model development and evaluation, and work closely with team members to integrate machine learning models into larger software systems.

Effective communication also plays a crucial role in troubleshooting and problem-solving. When issues arise, senior MLEs need to be able to clearly explain the underlying causes and propose appropriate solutions. They should be able to break down complex problems into smaller, manageable components and work collaboratively with team members to identify and address the root causes of issues.


While the rise of LLM APIs has undoubtedly made machine learning more accessible, it has also created a misconception that one can become a proficient MLE without acquiring a strong foundation in the core concepts. People should be cautious of the hype surrounding these APIs and recognize the importance of investing time and effort in learning the fundamentals of machine learning. Companies promoting their APIs should also take responsibility for emphasizing the value of foundational knowledge and not solely focusing on the ease of use.

As the field of machine learning continues to evolve, it is crucial for aspiring MLEs to build a solid understanding of core concepts in Deep Learning. Only then can they truly harness the power of LLMs and other advanced ML techniques to build robust, customizable, and innovative solutions. By combining the convenience of APIs with a deep understanding of the underlying principles, MLEs can unlock their full potential and make meaningful contributions to the field of machine learning.