Advice

Career

• Software Engineer

ML Pitfalls

• Personnel interdependence and complexity. Too many cooks and they are not in sync
• Lack of stakeholder demand for solution and willingness to change the current process
• Scope creep
• Technical gap between project and personnel
• Access to clean/ quality data
• A solution in search of a problem. ML is a solution but we are dealing with quantitative problems that may not need ML
• Product owner sets the stage and the vision, but the solution design must be driven by engineering
• No product management
• Gap in skills/ roles; cloud eng, data eng, mlops, sme, end-user
• Poor communication between the parties involved

Technical Presentations for a General Audience

• Have a slide or two to introduce yourself and your team. You can consider this talk as an opportunity to do marketing for you and your team; sell us on the notion that you are doing good work and inform us as to when the rest of the company should think to reach out to you
• What are the 2 or 3 big points that you want the audience to think about after the talk is over. If they were to summarize your presentation to someone else, what would you want them to say? I strongly recommend that you write these down for yourself and as you put slides together, ask yourself “is this content in service of the main points I want my audience to take away?”
• Assume that the audience is not familiar with your domain; there are audience members from across the whole company. We have members coming from very different departments; finance, supply chain, HR, engineering, etc. So keep this in mind when introducing the problem.
• Given that your domain may be new to some people, also try to motivate the project; why is this an important problem to solve? What’s the bigger picture? What will deploying this solution into production mean for the business?
• Why did you choose the approach you did? Compare and contrast your methods against other “popular” approaches
• Consider walking us through a specific example of the type of question your model is trying to answer; e.g. if you are classifying images, show us an example and explain what makes this problem different than other similar ones.
• Both the business value and the technical details are important. Aim the talk for a data scientist with about a year of work experience. So they should know the basics of DS and ML but if you can connect what you could read in an ML textbook to how you were able to implement a solution, that would be very instructive to this audience. However, the business impact is also significant; it would be great to convey that we want to solve valuable problems that affect the business, and therefore communicate the challenges of what it took/takes to go from a solution that is great on paper (or on your laptop) to a solution that can be embedded within a business process.
• Another way to think about it: if a junior engineer was going to try to replicate what you have done, what do they need to know? What were the technical and business challenges, and how did you tackle them? e.g. if you had to spend a lot of time working with stakeholders because the previous solution was deeply embedded within the business process and you needed to get significant buy-in stakeholders, then this is an important lesson to hear about.