Feb 2025 – Present · 1 yr 4 mos · San Francisco Bay Area

• Part of Menlo VC Fellows program for 2025.

Dec 2021 – Mar 2025 · 3 yrs 3 mos · Mountain View, California, United States · On-site

• Technical lead for Google Gemini LLM agentic platform. Led the scoping, design and implementation for integration of multiple extensions/tools with agents platform. Collaborated with SFT data curation, evaluation and metrics teams for quality improvement of first party and third party agents.
• Worked on Google Assistant discovery team to grow user engagement (DAU, L7) and retention (WoW) for multiple channels and devices impacting 100M+ users. Received a "Transformative Impact" rating which is for top 4% of Google employees.

Oct 2020 – Dec 2023 · 3 yrs 2 mos · San Francisco Bay Area

• Angel Investor in 17 early stage companies since late 2020. Out of this portfolio, 4 have exited by acquisition and 2 have received follow on investment including one by Andressen Horowitz.

Dec 2019 – Oct 2021 · 1 yr 10 mos · San Francisco Bay Area

• Built the engineering team for VinBrain from 3 engineers to 5 teams of total 35 engineers. Led the teams responsible for product development including ML training/deployment pipelines, backend services and web and mobile frontends from scratch. VinBrain's flagship product, DrAid, won the ACM SIGAI Industry Award under my leadership.
• Also led several partnerships including research partnership with Stanford University (Dr Andrew Ng's group), US medical domain datasets purchase agreements and USFDA 510k clearance consultants.

Sep 2018 – Nov 2019 · 1 yr 2 mos · Mountain View, California, United States · On-site

• Launched the AutoML NLP products for Google cloud platform. Created pre-trained and custom models for NLP problems like text classification, sentiment analysis and entity extraction using deep neural nets and AutoML.

Nov 2011 – Aug 2018 · 6 yrs 9 mos

• Led multiple projects at Microsoft's Speech Recognition and Natural Language teams. Re architected the Automatic Speech Recognition (ASR) pipelines to use Microsoft's CNTK (similar to Tensorflow, Pytorch) as ML framework and scale up the training pipelines and data parallel GPU training to handle 10x larger training data.
• Worked on some of the earliest production launches of large model distillation and noisy data augmentation techniques for speech recognition. Also was the lead developer for implementing large scale multi GPU SGD training in CNTK. Experimented with multiple recurrent (LSTM, GRU, RNNs) and convolutional(CNN) model types to train high quality models offline while maintaining low latency serving for speech use cases.
• Received a US patent based on this work.
• Previously led multiple projects for Query Rewriting team for Bing web search.

Feb 2010 – Dec 2011 · 1 yr 10 mos

• Worked on outdoor WiFi localization techniques using crowd sourced data collected from Windows Phone 7 users.
• Worked on creating an API that ensures location privacy on mobile phones.
• Received 2 US patents based on the work.

May 2008 – Jul 2008 · 2 mos

• Worked on indoor wireless localization techniques with Networks Research group at Microsoft Research, Redmond
• Filed a US patent based on the work.