What we talked about
Nithin Thekkupadam Narayanan joins the show to unpack the craft of product management at scale:how to keep a mature platform stable while innovating fast. A Senior Principal PM (ex-AWS, ex-Remitly and an MIT SDM alum), Nithin shares pragmatic lessons from event-driven systems, cost/speed trade-offs, and the fast-moving AI landscape.
Show notes
Nithin Thekkupadam Narayanan tells a story that most product managers would rather not admit: he launched a feature that would save customers 40% on storage costs, and almost nobody adopted it. The lesson he took away is that cost savings are not always the king, and that understanding the real priorities of mission-critical users is far harder than reading the market signals.
What we covered
- Event-driven architectures are becoming the backbone of AI workloads because old data is stale data. Nithin’s product, Oracle’s Transactional Event Queue, evolved from writing to disk (slower) to writing directly to RAM (faster), and then to natively supporting vector data types so that AI workloads do not lose time on format conversions before accessing data.
- The speed-cost tradeoff is not theoretical. Nithin described a small company that ran performance tests on Kafka, forgot to shut down the infrastructure, and received an end-of-month bill that exceeded their entire annual IT spend. Every design decision about speed involves a corresponding decision about what you are willing to pay for it.
- Launching a managed Kafka storage feature that cut costs by 40% produced unexpectedly slow adoption. The reason: when you are running mission-critical workloads, “if it’s not broken, don’t touch it” takes priority over saving money. A year after launch, many customers who said it sounded great still had not activated it.
- On feature prioritization, Nithin’s most consistent method is quantifying the signal: bucket similar feature requests from different customers into themes, count how many customers share the underlying need, and use that data to make the case to internal stakeholders. The biggest failure mode is being in limbo, a decision that is neither yes nor no is worse for the customer than a firm no with a clear reason.
- Oracle’s Converged Database consolidates relational, document (JSON), graph, and vector data types into a single system, with the Transactional Event Queue built directly in. This eliminates the need to manage a separate message broker alongside multiple specialized databases, reducing an architecture that might require five or six systems for a single use case down to one.
- The EVP at Oracle framed it this way: “AI is not going to replace people, but it will replace people who do not know AI.” Nithin described using deep-research AI workflows, not just Q&A with an LLM, to synthesize market data and build the internal case for new product investments, compressing a process that would previously have taken weeks.
About Nithin
Nithin Thekkupadam Narayanan is a Senior Principal Product Manager at Oracle, where he leads the Transactional Event Queue product. He previously held product roles at AWS and Remitly, and holds a degree from MIT’s System Design and Management program.
- LinkedIn: http://www.linkedin.com/in/nithin-thekkupadam-narayanan-0bbb4384
- Website: https://oracle.com
Episode 78 of the PreVetted Podcast.