ML Team Leader Q&A Part 2 - ML Career Path (5min read) | Life, The Universe, and Everything

type

status

date

slug

summary

🎯 Your AI/ML Career Options - A Brief Guide 🎓👩‍💻

Scientist in Academia 🏛: Propel the frontier of AI/ML, publish original research, and enjoy the liberty to dictate your research direction. But remember, you'll need to secure grants, publish regularly, and may not benefit directly from commercial applications of your work. 📚🔬

Applied Scientist in Industry 🏢: Use AI/ML to solve real business problems in a team setting. You'll have access to tons of real-world data and resources, and your work will directly impact business improvements. However, chances for original research and external publication could be limited. 📈💡

Engineer in a Startup 🚀: In this fast-paced role, you'll implement AI/ML systems quickly based on customer feedback. You'll need to be creative with limited resources and might have to wear many hats including ML engineering, MLOps, and tech support. Career advancement may be limited in this environment. 🌐👨‍💻

Engineer in a Large Company 🌇: In a larger company, you'll convert AI/ML research and prototypes into real systems. Expect defined roles and a clear career path with chances to work on large-scale systems. As a trade-off, until you reach senior levels, your influence over technical direction might be restricted. 🏭⚙️

Manager in a Large Company 🏦: As a manager, you'll define the AI/ML vision for your teams and lead its delivery. Your responsibilities span strategic planning, risk mitigation, product delivery, operations, career development, performance management, and hiring. In this role, you'll be the glue holding science and engineering together. 📊💼

Explore these options to find the AI/ML career path that suits your interests and aspirations! 🚀💫🧠

⛩️ Machine learning engineering team vs applied science team?

I have alway been asked to give recommendations on which team they should choose. Here are some of high level comments on both types of team.

MLOps engineers : Scientist = 2:1, a junior engineer for execution and a senior engineer to train junior and work with scientist

Until the science work is fully peer-reviewed, there is no engineering work scheduled

MLOps should has scientist and end customer in mind

How to lead a ML team:

doing planning probabilistically

attempt a portfolios of approaches

measure projects based on inputs, not results

having research and engineers work together

get end-to-end pipelines quickly and demonstrate quick wins

educate leadership on the timeline of uncertainty

🍡 Startup or not?

🚀🏢 The tech industry offers an intriguing dichotomy between the high-octane world of AI/ML startups and the well-oiled machinery of corporate giants. This prompts a career-defining question: Do you plunge into the feverish intensity of startups or navigate the steady currents of established companies?

🕒Speed & Adaptability: AI/ML startups operate on a landscape of quicksand, constantly shifting and adapting. Driven by the ticking clock of finite funds, startups must sprint to generate immediate outcomes and enduring solutions 📊. This setting encourages nimbleness and swift course corrections in response to feedback. On the other hand, large tech corporations proceed at a steadier pace, guided by meticulous planning and execution. Yet, in a twist of irony, these corporate behemoths might outpace the agile startups by deploying a concentrated force of people and resources to tackle an issue—cue the fable of the speedy hare being outdone by the patient tortoise! 🐢🐇

⚙️Organizational Structure: The DNA of a startup embodies a cohesive team passionately committed to a single product. There is complete transparency as each member is intimately aware of the team's collective efforts. Large corporations, however, resemble a sophisticated mechanism, with numerous interconnected teams and multiple products. These individual teams are goal-centric, concurrently working on several products. The key challenge? Ensuring alignment with the overarching strategy while maintaining a laser focus on their respective tasks.

🏦Resources: Startups cultivate a culture of thrift as they endeavor to accomplish lofty goals with shoestring budgets. This necessitates the birth of innovative methods and improvisations to build systems amidst data and compute constraints. Conversely, large corporations appear to be a treasure trove of resources, eclipsing the limited means of startups. However, this abundance brings its own set of challenges: the imperative to manage the surplus effectively and the need to defend budgets. Inter-team competition for projects and resources could also come into play.

🌱Career Development: A stint at a startup offers a buffet of roles ranging from tech leadership to sales engineering. The flat structure fosters autonomy and places less emphasis on formal career progression. On the flip side, the corporate career path has clear role definitions and hierarchies, providing greater focus on career advancement and promotions. Here, roles are more specialized but offer invaluable mentorship from experienced professionals.

While embarking on a startup journey can be akin to navigating a minefield (given that approximately 90% of startups fail to offer a positive return to investors 💔), I staunchly believe in the merits of testing these waters at least once. This enriching experience not only provides a newfound appreciation for the resource bounty at larger companies but also nurtures an audacity to think unconventionally and question the established norms. So, whether you're drawn to the adrenaline rush of startups or the paced endurance run of large companies, each presents a unique roadmap to explore the exhilarating topography of the AI/ML industry. 🗺️

🍣 Lead a MLOps team?

🚀 MLOps' role is to transform AI/ML research into functional production systems. It's like a bridge connecting the world of academia to the real world. MLOps engineers need to work closely with the research scientists, sort of like an interpreter who converts complex scientific findings and ideas into real-world applications. Hence, these engineers need to be comfortable with ML tools and methods and capable of decrypting scientific research. Just as you would not hire a translator who is afraid of foreign languages, you wouldn't want an MLOps engineer who fears equations!

👥 When designing MLOps systems, think of it as a two-sided market: one side being your research scientists and the other, the end-users. The ML system should provide an easy-to-use interface for scientists to experiment, tune, analyse, and diagnose without confusing the end-users. It's similar to a well-organized workshop where the mechanic can access all the tools easily without bothering the customer who's waiting in the lounge.

🧩 Ownership is key in the MLOps team, from the offline model training stage to the online inference stage. This is crucial because every minute alteration in the models has to be accurately mirrored during inference. If different teams handle these stages, you'll spend a lot of time playing the synchronisation game, and even slight differences can lead to training/serving skew impacting accuracy. It's like a relay race; if the baton isn't passed smoothly, the entire team's performance could be affected.

⚖️ Don't try to fuse your MLOps and science teams too tightly or apply rigid software management techniques to science. Experiments often need extra review time, and failure is common in science. If your MLOps engineers are always waiting for the scientists to finish up, it can lead to frustration. Similarly, forcing strict schedules on the research and experiments could be detrimental to the quality. The engineering work should only commence once the science has been thoroughly peer-reviewed, much like how a director waits for the screenplay to be perfect before beginning the shooting.

👷‍♀️ MLOps needs a significant amount of engineering effort. Imagine a ratio of 2:1 MLOps engineers to scientists. The engineers will have to maintain the existing ML system and aid scientists in developing the next one. You'll need a blend of senior engineers to tackle complex ML challenges and junior engineers for the routine tasks, similar to a well-coordinated soccer team. Senior engineers act as mentors for the junior ones, ensuring a sustainable team lifecycle. 🔄

💡

Let me know what else you want to hear about this topic.