Search & Intelligence

Google AI Research

Biometric Authentication / Pixel Face Unlock

Google-internal Equivalent of LangChain Framework

Parameter Efficient Fine-Tuning & Alignment for Generative AI (on-device)

Alexa Conversations

• Built a pipeline to facilitate finetuning (transfer learning) of a pretrained bi-LSTM-CRF model for NLU (intent and entity recognition) with Docker containers to fit inside existing workflow - https://amzn.to/2GLPBl1

• Productionized end-to-end inference stack running Intent Classification and NER for dialog system

• Designing POMDP dialog simulator that generates dialogs via self-play between a seeker and provider agent - paper accepted to NeurIPS 2020: Human in the Loop Dialogue Systems Workshop

• Derived Probabilistic Context-Free Grammars from utterances using Max-Likelihood Estimate

FireOS

• Wrote scripts and keys to cryptographically sign customer-facing code, such as Android APKs, Windows execs, and JAR files, to confirm the authenticity of the software author.

Multitask learning for medical images (radiographs)

Developing methods of feature extraction from images of tumors, and working on developing neural nets as models to train data of images on to accurately detect cancer in patients from CT and MR images.

Developed a serverless processing pipeline using Amazon Web Services Lambda, Simple Queue Service, and S3 that allows users to input images, preprocess them, and either train a model on given labels or predict a label with an existing model.