Working on LLMs - Gemini: https://arxiv.org/abs/2312.11805

• Technical lead of machine learning and search.• Designed AI services and a machine learning platform for Box.• Increased search relevance CTR (click-through rate) by over 30% and NDCG by over 20%.• Re-architected Box’s distributed search infrastructure across many data centers to increase availability from one to four nines. Experience with Spark, HBase, Solr, PHP, K8s, TensorFlow. • Presented a search specific sharding design at the Activate conference in Montreal. • Solved a large-scale learning problem by creating a click & event prediction system in Spark that increased CTR by 15%. • Coded pipelines for deep learning reranking, spark ML, A/B test reporting, metrics and back testing to create a customer feedback driven flywheel of perpetual customer data driven satisfaction.• Led multiple engineering wide presentations to evangelize machine learning paradigms.• Mentored in career and technical aspects which led to multiple colleagues to up level.• Led teams for continuous delivery of ML features while extensively documenting design with multi-year visions.

• Designed and implemented a document extraction pipeline for automating the mortgage process. • Used OpenCV for image processing which reduced Google’s OCR error count on W2s by 4x. • Created a backend that ran on GCE, Kubernetes, Terraform, Docker, NGINX, Django.

• Created backend services in Python and Go on a global scale, such as the Uber feed and tools for operations such as airport geofencing for better pickup point management.

Instagloss.com• Created a reading tool to help understand complex content using machine learning. • Featured variable summarization and fast entity extraction.

• Member of Microsoft’s Bing relevance team; particular feature teams being the speller, search results ranking, autosuggestions, query understanding, query alterations and concept recognition. • Released advances to the Bing infrastructure that resulted in machine reductions of over 83% with favorable user metrics and feedback. • Processed petabytes of data through a map-reduce framework to discover new patterns in search behavior. Applied data mining to yield session-based query transition and language models.• Shipped: Three different spellers (neural net, lambda rank decision tree, statistical machine translation), semantic search index expansions, the first iteration of the suggest as you type feature and data-mining tools written in C# and python.

• Pow hound & teacher of 1000s.