← All speakers

Bio, Work & Ideas

Bala Ramdoss

Conference affiliation: Amazon · 2026

Bala Ramdoss is a senior software engineer and mobile technical lead at Amazon working on Amazon Lens, the camera-based shopping system that identifies products from images, screenshots, and barcodes. His experience building customer-facing mobile applications shapes a pointed approach to AI products: an agent’s answer becomes useful only when an interface turns it into an action.

Ramdoss’s earlier work included Android-based fall-detection research combining motion sensors with mobile alerts. He studied mobile development and the Internet of Things at Santa Clara University, worked as a graduate teaching assistant, and wrote about the practical differences between iOS simulators and Android emulators. At Amazon, he developed camera-driven shopping experiences including Lens and Lens Live, bringing together Android engineering, visual search, and production mobile interfaces. His technical leadership work also addresses incremental delivery, cross-team alignment, and communicating engineering trade-offs.

Building interfaces that complete the job

Ramdoss treats generative UI as product infrastructure: a restaurant recommendation should become a booking interface, not another research task. His approach to actionable AI interfaces centers on three production patterns:

  • Version-aware rendering contracts: Give models a catalog of components each installed app version actually supports, preventing generated interfaces from calling unavailable native elements.
  • Progressive structured-UI streaming: Deliver useful interface components incrementally and optimize for the first meaningful interaction. Lens Live keeps shoppers engaged with an interactive camera while results arrive.
  • Backend-for-Frontend orchestration: Translate model intent into familiar Android or iOS components, attach actions and deep links, preserve context, and handle incompatible output safely.

He favors controlled or declarative component systems, including the design approach behind Google’s A2UI, over unrestricted generated screens. The model chooses what the customer needs; the application determines how that experience can ship safely.

Read the topics behind these talks

1 conference talk

Key ideas

Scroll to read ↓

A capable agent can find the right information and still leave the user doing the work. Typed UI contracts, streaming states and a backend for frontend turn that output into an usable interface.

  • The reservation is still unfinished
    0:01 ↗
  • Describe the interface as data
    1:56 ↗
  • Design for clients you cannot promptly patch
    5:06 ↗
  • Constrain component selection before generation
    6:35 ↗
  • Stream useful component states
    8:42 ↗
  • Give the wait a useful interaction
    10:13 ↗
  • Turn UI intent into actionable native components
    11:18 ↗

References