429 Problems: Scaling Systems Without Breaking Everything

We’ll explore the real world tradeoffs of parallel processing through a case study of an AI pipeline that processes and analyzes large volumes of user-submitted political opinions. The system performs moderation, embedding generation, and downstream analysis using a distributed, event-driven architecture built on AWS.

In this session attendees will learn how to design scalable, resilient pipelines using parallel processing while avoiding common pitfalls like rate limits, throttling, and cascading failures. They’ll leave with practical patterns for building fault-tolerant systems, balancing throughput with external constraints, and handling partial failures gracefully.

We’ll break down how parallelism improves throughput and responsiveness, while also examining the less obvious challenges it introduces, such as rate limiting, back pressure, partial failures, and data consistency.

While the case study uses AWS services, the architectural principles are cloud-agnostic and applicable to any distributed system handling high-throughput workloads.

***

See you at the next DSM AI Event:
5:30 - 6:00 - Socializing/Networking
6:00 - Speaker begins