Pranav Prabhakar is an engineering leader with expertise in building scalable systems, deploying machine learning in production, and leading high-performing remote teams. As Co-founder & CTO of MiStay, he architected the platform from the ground up. Later, as Engineering Manager at ManyPets, he drove ML-driven claims automation, backend architecture redesign, and asynchronous collaboration initiatives. His work combines deep technical expertise with a focus on achieving measurable business outcomes in complex, regulated domains.

Analytics Insights reached out to Pranav to discover how he became an engineering leader, reflects on the nuances of asynchronous work, and shares his perspective on the future of ML and automation.

1. You have had a unique career path, from co-founding MiStay to leading engineering and ML initiatives at another company. Can you walk us through your journey and what shaped your focus as a technologist and leader?

I started my career by co-founding MiStay - a travel-tech startup that pioneered slot-based hotel booking in India. As a founder, I juggled many roles - building the product from scratch, setting up integrations with hotels, and scaling the backend systems to handle real-world constraints. This experience has shown me the importance of reaching a balance between fast experimentation and long-term scalability.

Later, when I moved into a larger organisation, I carried forward that entrepreneurial mindset but applied it to more complex, regulated domains. For example, I had the opportunity to lead engineering teams and coordinate machine learning initiatives in claims automation. In this case, I refocused from just "building things fast" to securing reliability, compliance, and measurable business outcomes.

In my opinion, the recipe for success as a leader is using technology not just to solve problems, but to design systems that can scale, adapt, and create lasting impact in both startups and mature companies.

2. Many early-stage startups face scaling bottlenecks. How did you strike a balance between scrappiness and building a scalable foundation at MiStay?

I believe this balance is what separates startups that can grow sustainably from those that burn out under technical debt.

At MiStay, every decision was a trade-off between speed and sustainability. Initially, we optimised for scrappiness - getting features live quickly so we could test whether customers actually wanted or needed them. However, we also identified areas where cutting corners would create long-term pain. For example, payment flows and booking availability were non-negotiable, as they had to be reliable from the outset.

The balance came from consciously deciding where to invest and where to cut back. Not everything needs enterprise-grade engineering at the beginning, but certain "core loops" of the business do. Over time, we evolved lightweight yet scalable patterns - such as modularising integrations and building monitoring proactively - which helped us scale without a complete rewrite.

3. You have led ML initiatives in claims automation. What are the most practical challenges in moving from proof-of-concept to production when it comes to ML?

The biggest challenge is not model accuracy, but everything surrounding the model. In proofs of concept, you can show great results on curated datasets. However, in production, real-world noise, missing data, compliance constraints, and integration complexities become significant factors that must be addressed.

Some of the most practical challenges I have observed in deploying ML systems include issues of data quality and drift, where models can silently fail if the underlying data changes, making monitoring and retraining loops essential. Another major hurdle is integration with legacy systems, since machine learning must fit into existing operational workflows and APIs rather than starting from scratch. Reliability and explainability are also critical: stakeholders do not just want accurate predictions. Instead, they need to understand and trust them, particularly in regulated industries. Finally, the operational overhead of deploying, versioning, and rolling back models must be optimised to be as consistent as standard code deployments.

4. You recently introduced "async ways of working". What worked, what did not, and what would you advise leaders struggling with remote collaboration fatigue?

Async ways of working were our response to the challenges of a fully remote team. What worked particularly well was documenting decisions, using structured templates, and converting routine updates into written form. This freed up meeting time for real discussions rather than status checks. It also created a culture of transparency - people could align on context without needing a call.

What did not work initially was expecting everyone to adopt async habits immediately. Some team members missed the spontaneity of synchronous chats, and without clear norms, async communication risked becoming fragmented. We learned that async is not "no meetings". In fact, it is about being intentional with sync time.

For leaders, my advice would be to invest in shared artefacts, such as documentation, boards, and recordings, to keep knowledge accessible and transparent. At the same time, teams should define clear response expectations, recognising that not everything requires immediate attention. Finally, to maintain human connection, regular yet focused sync sessions help achieve a balance between efficiency and purposeful collaboration.

Async is not a silver bullet, but when combined with intentional sync, it reduces fatigue and makes remote work more sustainable.

5. If you had to bet on one area where ML and automation will have the most business impact in the next five years, what would it be?

I see the most significant impact coming from intelligent process automation - using ML to optimise the repetitive, low-value tasks that still consume a surprising amount of human time. Advances in natural language models, OCR, and workflow orchestration now make it possible to automate tasks that were previously overlooked due to rigid systems. That shift can free people to focus on judgment, creativity, and human interaction, rather than manual "glue work."

Nevertheless, the story is not just about replacement. Most jobs are a mix of routine and non-routine tasks, and automation typically displaces the former while amplifying the latter. Recent evidence suggests that ML will reshape jobs rather than eliminate them: humans will increasingly supervise, guide, and refine automated systems. At the same time, organisations will need to rethink job design, invest in reskilling, and address the ethical and regulatory questions that come with automation at scale.

In short, the next five years will bring more innovative models, as well as strategic transformation in how work itself is structured, with ML embedded into everyday processes and humans moving into higher-value, more human-centric roles.