Shrestha Basu Mallick

Shrestha Basu Mallick was a Google DeepMind Group Product Manager and lead for quality at Gemini Enterprise at Google Cloud Next 2026. Her earlier responsibilities included product leadership for Gemini API, Google AI Studio and AI assistance across Google’s coding surfaces. Her public work explains practical choices in bringing models to developers: when to retrieve outside information, how to expose structured outputs, and how to make multimodal applications responsive.[2][4][3][5]

Before her enterprise-quality role, Basu Mallick helped lead Gemini’s developer products. In an October 2024 interview with Nathan Labenz, alongside Logan Kilpatrick, she described three priorities for the team’s roadmap: learning from other model providers, offering capabilities Google could uniquely supply, and improving the everyday API experience. She placed error codes, support, rate limits and affordability in that third category, and said developer feedback was guiding improvements to initially released features such as structured outputs and context caching.[2][5]

Basu Mallick and Kilpatrick coauthored the October 31, 2024 announcement of Grounding with Google Search for Gemini API and AI Studio. The service retrieved relevant search information at query time and supplied it to the model; responses included supporting links and Search Suggestions. Dynamic retrieval added a second layer of control: a prediction score estimated whether a prompt would benefit from grounding, and developers could set the threshold for invoking search. The announcement recommended testing thresholds for each application because retrieval added cost and latency.[4]

For Basu Mallick, grounding also raised a question about what kind of answer a user wanted. In the October interview, she recalled preferring a search-grounded answer to a question about Mars’s capital that discussed fictional capitals in science-fiction books; she said Kilpatrick preferred the direct answer that Mars has no capital. She explained source links as serving both publisher referrals and users’ ability to validate an answer or explore its nuances.[5]

She also discussed the team’s decision to make structured-output fields optional unless marked required. Basu Mallick framed this as following the team’s interpretation of developers’ expectations about JSON, while acknowledging the reasoning behind another provider’s choice to require fields and remaining open to feedback. Her emphasis on affordability came with a qualification: she argued that developers were justified in watching for hidden costs from tool use and code execution, even as inference costs declined.[5]

In July 2025, Basu Mallick coauthored the GenAI Processors announcement with engineers Andre Elisseeff, Alexey Guseynov and Oskar Bunyan. The open-source Python library gave input handling, preprocessing, model calls and output processing a consistent Processor interface. Text, audio chunks and image frames flowed through pipelines as asynchronous ProcessorPart streams. The library automatically ran eligible operations concurrently while preserving output ordering relative to the input, with the aim of reducing time to first token.[3]

The announcement illustrated this approach with a live agent combining microphone and camera input, Gemini Live API processing, and audio playback that paused when the user spoke. A second example connected speech recognition, a text-based model and speech synthesis. The coauthors argued that processing data as soon as it became available could improve responsiveness even for turn-based applications; they did not supply a quantified latency improvement.[3]

Basu Mallick also participated in the June 2025 Latent Space preview of Gemini and real-time voice AI, recorded with TWiML AI’s Sam Charrington at Google I/O. The preview listed her joint AI Engineer World’s Fair workshop and talk with Kwindla Hultman Kramer on real-time workflows using Gemini Live API and Pipecat.[1]

Before her Gemini work, Basu Mallick’s career included scientific training and product work in materials discovery. She holds a PhD in Applied Physics from Stanford. Her first Alphabet role was Head of Product for a materials-discovery platform at X, the Moonshot Factory; that platform later became a separate startup. Before Google, she held product and strategy roles at Salesforce Einstein, McKinsey and Docusign.[2]