Startup Life: Unscripted #39 with Karan Acharya, Machine Learning Engineer at Guardrails AI

Presented by

Startup Life: Unscripted is a TNG Media newsletter, as part of The Nudge Group, where we feature candid conversations with startup operators about their career journeys and experiences. If you received this email as a forward, you can read all our past interviews and subscribe right here.

Welcome back to Startup Life: Unscripted! Today, we sit down with Karan Acharya, a Machine Learning Engineer at Guardrails AI. Moving from academia at Indiana University Bloomington to the bustling startup scene has given Karan a unique blend of experiences and insights, which he’s eager to share.

Key interview takeaways:

📚 Study to Startup: Karan breaks down how his deep dive into Applied Algorithms and Machine Learning at university became his toolkit for startup success.

🔧 Theory to Action: From classroom concepts to crafting an object detection model at Wobot.ai, Karan reveals the thrill of turning ideas into impactful technologies.

🛠️ Startup Skillset: He shares the essential blend of multitasking, collaboration, and communication skills that fuel his daily success in the fast-paced startup environment.

🌱 Learning on the Fly: Learn about Karan’s strategy for juggling diverse projects and his relentless pursuit of new knowledge in the ever-evolving tech landscape.

Got a minute? Share this newsletter with your network and help us inspire more individuals with real-life startup stories and valuable insights.

Karan, transitioning from academic studies to a fast-paced startup environment at Guardrails AI must have been a significant shift. How did your experiences at university and as an intern prepare you for the dynamic nature of startup life? 

The jump from Indiana University Bloomington to Guardrails AI was a big shift! But the experiences I gained during my academic journey and internships proved valuable. 

For example, taking up courses like Applied Algorithms, Applied Machine Learning, Elements of AI, Data Mining and Explainable AI gave me a solid foundation in research and development.

Additionally, I took up Independent Study courses and worked as a Research Assistant in the field of Interpretable ML for Computer Vision. I consider myself fortunate to have had amazing professors at Indiana University, who inspired genuine curiosity and challenged us to translate high-level, complex problems into efficient end-to-end solutions through great discussions and assignments. 

During my time as a Graduate Teaching Assistant for two graduate-level ML courses, I had to manage a lot of moving parts – course data, student projects, and individual needs. This experience honed my ability to juggle multiple tasks and deliver projects on time, which is a must-have skill in a startup. 

My internship at Wobot.ai was particularly valuable. It wasn't just about building models; it was about deploying them, designing real-world use cases by automating tasks and building end-to-end pipelines. This experience directly translates to the kind of work I do now at Guardrails AI, building and deploying solutions that make a real difference. 

Finally - conducting research, brainstorming approaches, conducting experiments, helping write a research publication and delivering presentations helped me not only become a better software engineer but also develop strong communication skills. Assisting and mentoring students honed my ability to collaborate effectively.

All these skills are crucial for success in a startup environment where clear communication and teamwork are essential. Last, but not the least - always embracing a student mindset and thinking from a place of “learn-it-all”, instead of “know-it-all”, as popularised by Satya Nadella has helped immensely - and continues to do so. 

Your time as a Computer Vision Intern at Wobot.ai to now working at Guardrails AI sounds like you've had quite the hands-on experience in the field. What's been the most exciting project you've worked on so far, and what made it stand out for you? 

From Wobot.ai to Guardrails AI, I've been fortunate to dive right into practical applications of computer vision and machine learning in general. While all the projects have been rewarding, the one that stands out for me is the object detection model pipeline we built at Wobot.ai. 

Here's why it was so exciting: 

• Real-World Impact: It wasn't just about building a model; it addressed a real need – by detecting instances of compliance violation and ensuring whether rules were being followed. Seeing the model go from development to deployment was incredibly satisfying. 

• Full Project Lifecycle: I got to experience the entire project lifecycle right from data collection, pre-processing, augmentation, model training to deployment and back to data cleaning and training. This well-rounded experience solidified my understanding of how computer vision can be applied in the real world, across all stages - and what are the challenges at each step. 

• Team Collaboration: We had to work together effectively to achieve our goals, and that teamwork is a big part of why the project was such a success. This improved my research, communication, and presentation skills. 

• As a stepping stone: That project went through multiple iterations, and in some instances it was difficult to understand, and explain why the model predicted the way it predicted - even for the correct classes. This prompted me to dive deeper into understanding the challenges with ML models in the wild and learn more about the field of Interpretable ML and Explainable AI (XAI). This later manifested into getting a master’s degree with a research focus in that precise domain. 

Overall, this project at Wobot.ai gave me a chance to see the entire picture – from identifying a need to building and deploying a solution that made a tangible difference. It was a great springboard for the work I'm doing now at Guardrails AI.

With LLMs it became easier than ever before to use and deploy a Machine Learning model - but also raised a unique set of challenges that we're constantly learning and innovating. 

Startups often require a multifaceted skill set. Given your diverse interests from app development to machine learning, how do you prioritise learning and contributing across different areas? 

That's a great question! You're right, startups thrive on people who can wear multiple hats. Here's how I approach prioritising learning and contributing across different areas: 

• I constantly evaluate my skills and the needs of the company. If I am not familiar with a concept that I just heard during a meeting, or for a task that has been assigned to me, I try to first get an intuitive sense of what that concept is, what problem it solves, how it connects to what I already know and then dive into the specifics of how each thing within that thing works. I prefer a top-to-bottom approach when learning new things. 

• In a fast-paced environment, dedicated chunks of learning time can be scarce. So, I use a micro-learning approach, taking online courses, reading tutorials, or watching videos during downtime. I then try to apply this new knowledge to current projects. 

• My interests in app development and explainable AI complement my core skills. For example, my understanding of app development helps me grasp user interfaces and developer experiences for future feature updates, while the research ML knowledge lets me contribute to adding and improving our existing model-based validators. 

• I subscribe to and follow interesting people on LinkedIn and Twitter who share their own learnings or share resources to external articles that tackle problems and provide valuable, interesting insights into how users are using ML models (and LLMs) in production and what challenges and problems they face and their solutions. The goal here is: to keep myself exposed to new ideas and play the game of perspectives. 

• Last but not the least, the team at Guardrails AI is phenomenal! Everyone is very approachable and helpful, and I constantly ask a lot of questions and get my doubts solved.

Why are all your favorite newsletters switching to beehiiv?

It’s because the founding beehiiv team were all early Morning Brew employees who helped scale that newsletter to over 4 million daily subscribers.

Years of trial and error went into building the precise tools, dashboards, and analytics needed to accomplish that. And now every newsletter on beehiiv has access to the same winning formula.

So what exactly does beehiiv offer?

• World-class growth tools like the referral program and recommendation network

• Monetization via the beehiiv Ad Network and premium subscriptions (i.e. beehiiv helps you get paid)

• Seamless content creation with a sleek collaborative editor

• Best-in-class inbox deliverability of 99%

• Oh and it’s the most affordable by a mile…

Take your newsletter to the next level — get started for free.

You've mentioned your passion for diving deep into subjects you're curious about. How do you apply this approach to identifying and tackling new projects or technologies? 

My curiosity fuels my approach to identifying and tackling new projects or technologies! 

• I keep my eyes and ears open for interesting problems or emerging technologies. This might involve attending industry talks, listening to podcasts, or simply browsing online forums. When something sparks my curiosity, I start digging deeper. 

• Once I'm hooked on a topic, I dive in headfirst. I might take online courses, scour research papers, or even experiment with open-source projects. This "deep dive" lets me understand the core concepts and potential applications. 

• Here's where the fun part starts! I attempt to connect my new knowledge with my existing skills and what we want to achieve at the company. Can this new technology be used to improve our existing models? Could it address a challenge we haven't considered yet? Could this turn into an exciting use-case? This is where I brainstorm potential applications and project ideas. 

• Not every curious exploration leads to a full-blown project. I assess the feasibility of my ideas based on factors like time, resources, and potential impact. If it seems promising, I present it to the team. Sometimes, these discussions unearth even better ways to leverage the new technology. Other times, I realise it was just an exciting idea and may not be feasible in production - but that’s when I learn the most. 

• The learning process doesn't stop once a project starts. I continue to learn and refine my understanding as I work on the project. This iterative approach ensures I'm not just building something, but also expanding my knowledge base for future endeavours. 

So, my passion for diving deep keeps me curious and constantly learning. This, in turn, fuels my ability to identify and tackle new projects or technologies that can benefit both my own growth and the work we do at Guardrails AI. 

In our conversation, you mentioned a love for authors like Cal Newport and their philosophies on deep work and digital minimalism. How do you apply these principles to your daily routine, especially in a field as demanding as machine learning? 

You're right, I am a big fan of Cal Newport's work on deep work, digital minimalism and slow productivity. Machine learning and software engineering are demanding areas, and staying focused can be a challenge. Here's how I try to apply these principles to my daily routine: 

• Blocking Time: I schedule dedicated "deep work" blocks on my calendar. During these times, I silence notifications, turn off email and Slack, and focus entirely on a specific task. This allows me to enter a state of flow and make significant progress. 

• Batching Similar Tasks: Multitasking is a focus killer. I try to batch similar tasks together, like responding to emails and slack or attending meetings. This minimises context switching and allows me to maintain focus during deep work sessions. 

• Avoiding Distractions: Social media, news feeds, and irrelevant notifications can be rabbit holes. I silence notifications on my phone and laptop during focus blocks. 

• Curated Information Diet: I'm intentional about the information I consume. I follow relevant blogs and subscribe to AI newsletters, but I avoid endless scrolling and aim for high-quality, focused information. 

• Focused Learning: Instead of passively watching random tech talks and videos, I set specific learning goals. I might dedicate an hour to learning a new concept related to machine learning, using targeted online resources. 

• Digital Detox: Taking breaks from technology is crucial. I schedule time to disconnect completely, whether it's going for a walk or spending time on a non-digital hobby. 

• Focus on Quality Over Quantity: While "deep work" emphasises focused effort, Slow Productivity reminds us to do less but better. I apply this by prioritising the most impactful tasks for my deep work sessions. I schedule breaks throughout the day to avoid burnout and ensure my focus remains sharp during deep work sessions. 

These are just a few ways I try to incorporate the phenomenal ideas from author Cal Newport into my routine. It's a constant work in progress, but by being intentional about how I use my time and technology, I can stay focused, learn effectively, and make progress in the demanding field of machine learning. 

Final question before we wrap up. For someone just stepping into the vast world of machine learning, the road can seem quite daunting. Based on your own experiences, what advice would you give to those just starting out? Any must-read books or must-watch courses that you'd recommend? 

• Never stop learning - embrace a student mindset. The day you think you’re an expert at something - that’s when you have crossed the threshold and will no longer easily learn new ideas. It’s okay to be comfortable at something, but always strive to improve your skill. Be like a ML model - learn from your loss function and take the right steps in the right direction. 

• Learn how to unlearn. Learning new things requires one to be aware and identify what you currently know about a topic and be comfortable with the fact that what you knew yesterday is no longer true today and be willing to accept that and update what you know. As Adam Grant says in his book “Think Again” - think like a scientist, and not like a preacher or a prosecutor trying to preach, sell or attack other’s ideas. 

• Get better at identifying the low-hanging fruits and start doing something in that direction ASAP. Convert concept → code fast. Try to get to the lowest levels of abstraction. You don’t need to have everything figured out all at once. Most of the time, all that is required is the first step. Each step opens a different path you can go down, and you will figure it out once you’re there. Don't wait for the "perfect" project. Dive into online tutorials and platforms like Kaggle to get hands-on experience building and training models. Experiment with different datasets and algorithms. 

• Be okay with asking “dumb” questions. If you ask once, you’re a fool once; if you never ask, you’re a fool for lifetime. Ask more questions, get answers - and ask questions again. That’s how you’ll increase your circle of “known” unknowns. 

• Don’t chase the fancy tools, technologies, or titles. Adapt a craftsman mindset instead of a success mindset. Think more about what you can offer the world, not the other way around. Focus on getting better at a general skill, not specific sub-skills. If you want to be a better ML Engineer, or a Data Scientist or a Data Analyst, focus on getting better at software engineering and programming first - then worry about specific deltas required for each subset. 

• Reach out to people - not just to seek jobs; but to learn and understand something you’re curious about. Attend meetups, conferences, or online forums related to AI and machine learning. Connect with other enthusiasts and professionals. Sharing knowledge and experiences can be incredibly valuable. 

• AI is a vast field. Explore different areas like computer vision, natural language processing, or deep reinforcement learning. Find what sparks your curiosity and focus on developing expertise in that area. 

• Technical skills are crucial, but don't underestimate the importance of communication and collaboration. You'll need to explain complex concepts clearly and work effectively with others, especially in a fast-paced startup environment.

• Courses: 

  • Machine Learning by Andrew Ng 

  • Deep Learning Specialisation by deeplearning.ai 

  • Tensorflow in Practice by deeplearning.ai 

• Tutorials/Videos: 

  • Follow Andrej Karpathy on YouTube and Twitter 

  • Stanford CS230: Career Advice / Reading Research Papers

From the Startup Life team

And that's a wrap! We hope you've enjoyed this edition as much as we loved putting it together. Stay curious, keep learning, and above all, enjoy the rollercoaster ride that is Startup Life. Catch you in the next one! 👋 Not subscribed yet? Do it here and don't miss out! Subscribe Now.

🎙️ Check out Give It A Nudge, our video podcast that explores the world of start-ups, founders, venture capital, and private equity. Follow us on YouTube, Spotify, Apple Podcasts (or your preferred platform) to hear new episodes every fortnight.

📩 Want to get your brand in front of 14,000+ startup enthusiasts, operators, and potential candidates? For details on our sponsorship opportunities, simply hit reply and let’s chat!