Recap of Scala Meetup #7: Tubi Recommendation System Architecture, The Nature of Computation, and Reactive Streams in Large-Scale Scenarios

Scala Meetup #7 was held online with more than 1400 participants and featured three technical talks: the architecture and practice of Tubi's content recommendation system, the nature of computation, and reactive streams in large‑scale scenarios.

Junjun presented the Tubi recommendation system, describing an end‑to‑end architecture built on AWS that includes a Tensor gateway, MongoDB content store, user‑profile (TUS) service, ranking service, experiment platform, and a JSON‑based pipeline DSL that orchestrates model inference and feature retrieval.

Detailed diagrams illustrated the recommendation service gateway, the experiment platform’s namespace‑experiment‑group hierarchy, and the multi‑stage pipeline where each stage invokes specific model services, demonstrating the engineering challenges of maintaining a custom DSL and the need for strong backend expertise.

Gao Yuxiang explored the philosophical side of computation, quoting Wittgenstein, recommending the book "Gödel, Escher, Bach" and discussing category theory concepts such as categories, objects, and morphisms, as well as their relation to functional programming and Scala libraries like ZIO.

He Pǐn compared various JVM asynchronous models, shared the migration from Reactor/RxJava to Akka Streaming for a live‑streaming platform, and highlighted achievements like supporting 15 million concurrent clients and reducing resource usage by 40%.

The round‑table discussion addressed Scala's industrial adoption, community challenges, and the trade‑offs between functional paradigms and practical engineering constraints, emphasizing that language choice should serve business needs.

Audience feedback praised the depth of the talks, and the article provides links to recordings, slide downloads, and related Scala articles for further study.