Avinash Kumar

About Avinash Kumar

As an AI, engineering, and thought leader at Microsoft, I bring a blend of deep expertise…

Experience

1. Principal Data Scientist

   Microsoft

   Sep 2017 — Present

   Following are the highlights of some of my work: • Built a state-of-the-art model to remove inappropriate multilingual user queries and prefix suggestions in Bing Autosuggest. This model has helped in reducing inappropriate query leakage by X% and prefix leakage by Y%. • Incorporated a meta-learning approach to build a high-performant query classifier for detecting the adult intent of the user\'s query, which helped in reducing the adult leakage by X% in the Bing web search result. • Built a Generative RL-based model to prevent appearing low authority documents in Bing search results for defensive queries. The built model generates human-like synthetic queries that are highly semantically coherent (at least by XX%) w.r.t. input seed query. Most importantly, these synthetic queries are diverse enough to provide greater coverage (Y times) to prevent the low authority document from appearing in the Bing search result. • Conceptualized the problem of “right time to live migrate the VM in Azure” into multi-variant counter prediction task and built a high performant DNN model, which predicts “go” and “no-go” for Live migration of VM with a high precision of ~XX% and recall ~YY%. This model helps in dropping in brown-out time by ~Z%. • Built an unsupervised model to generate representative workload for a production-like test environment. The test environment, which is representative of the production environment, is helping in catching the faulty azure builds with very high precision (X%) and recall (Y%), which in turn has helped in reducing the interruptions across the azure clusters by Z unit. • Designed a common data Lake at the enterprise level, built data ingestion, workload monitoring frameworks, and applied platform/workload level optimization that helped various SPARK workloads to process terabytes of data by reducing the execution time and memory usage of the platform by and respectively.

Education

• Birla Institute of Technology and Science, Pilani

  Doctor of Philosophy (Ph.D.), Computer Science (Natural Language Processing, Deep Learning)

• Indian Institute of Information Technology

  Bachelor of Technology - BTech, Information Technology

Skills

• Xml
• Unix
• Databases
• Business Analysis
• Web Services
• Big Data
• Oracle
• Oozie
• Impala
• Sql
• Mapreduce
• Software Project Management
• Extract, Transform, Load (Etl)
• Shell Scripting
• Sdlc
• Hive
• Soa
• Avaloq
• Data Modeling
• Sqoop
• Hadoop
• Spark
• Kyc
• Anti Money Laundering
• Performance Tuning
• Apache Kafka
• Nltk
• Business Objects
• Nifi
• Uml
• Pl/Sql
• Informatica
• Service-Oriented Architecture (Soa)
• Natural Language Processing
• Database Design
• Data Mining
• Unix Shell Scripting
• Core Java
• Apache Pig
• Requirements Gathering