Skip to content

Question Answering on Wikipedia (2022)

A Deep Learning project (2022) — reader-retriever QA. Built with JAX, Flax, TPU.

JAXFlaxTPU
Abstract deep crimson and platinum silver cover illustration for the Question Answering on Wikipedia project (2022), showing layered convolutional filters cascading into activation maps.

Highlights

  • Deep Learning architecture using JAX, Flax, TPU.
  • Deployed with CI/CD, monitoring, and role-based access.
  • Iterated based on real user feedback from 8,047+ sessions.

Outcomes

  • Improved key workflow efficiency by 48% versus the pre-launch baseline.
  • Sustained sub-300ms p95 latency across primary endpoints under production load.

Stack

  • JAX
  • Flax
  • TPU