Akshay Arya

Currently lead the Smart Home Machine Learning Infrastructure & Data team. Key accomplishments include..

• Served as the cost champion for the Smart Home organization. Collaborated with 2 external organizations and partner teams across 3yrs to lead optimizations to ML infrastructure and Data ETL pipelines. Achieved cost-efficiency gains of >$10MM in annual savings run-rate.
• Collaborated with partner teams to manage delivery of several engineering, IOT sensor fusion and ML modeling post-launch improvements for Hunches Automatic Actions based yielding reduction in defect rate of 434 BPS .
• Managed engineering systems for the critical “Did You Mean” (DYM) and “Turnless” DYM experience in partnership with science and product teams. The systems are responsible for voice defect reduction for Smart Home by 45% , with >8MM/wk voice interactions recovered across 15 locales internationally.
• As the data custodian for Smart Home customer data, lead the development of several audits & controls around sensitive data management in collaboration with legal, security engineers and partner teams. Spearheaded the infrastructure-as-code direction for the development of controls, leading to widespread adoption across the organization and saving ~48 SDE weeks worth of effort for the organization.
• Managed over 11 SDEs including Sr. SDEs. Coached team through 7 promotions. Facilitated transition to Applied Scientist for interested MLEs. Experience managing out under-performers.
• Partner to the Amazon Military Apprenticeship Program. Coached a military apprentice from first line of “Hello World” to a full-time SDE in 6 months. Coached him through another promotion to mid-level SDE.

• Chief system architect of Hunches Automatic Actions - a cutting-edge NRT smart home event-driven inference engine with tight SLAs around data consistency, user perceived latency and precision. Achieved alignment on system architecture from multiple stakeholders including engineers, PMs and Sr. staff engineers across 3 organizations and 10 teams.
• Served as tech-lead for a team of 7 engineers through development of Hunches Automatic Actions in 8 months from architecture alignment. Dived head first into solving for multiple roadblocks and ideated controls around protecting sensitive customer data.
• Authored the launch-plan document for the rollout and received approval from all stakeholders including Sr Staff engineers. Directed the phased dialup across 3 organizations, 10 teams for Lighting, HVAC, smart water heaters and robotic vacuums. Launched to 2MM+ customers with no major issues while meeting the high bar for precision
• Successfully filed for 1 patent for the technology developed.

Led machine learning projects within Alexa from ideation to launch, followed by experiment based improvements and expansion. Selected achievements ..

• "Did You Mean?" : Led the project from ideation to launch. Architected and implemented the infrastructure to serve ML inference including but not limited to a Deep Neural linguistic model at Alexa scale. Also designed and implemented the feedback loop pipeline for continuous learning. Setup the experimental framework and lead the ideation and expansion of the feature to all I18N locales including non-english ones.
• Hunches : Expanded the ML inference pipeline for complex models for "Did you mean" to serve Hunches use cases. On boarded some of the complex models for Hunches to my platform and later advised on the refactor of the entire Hunches model serving infrastructure to my platform.
• Pseudonymization : Invented and implemented a psuedonymization solution using k-anonymization for the Alexa, Smart Home Machine Learning platform to protect sensitive customer information.
• Successfully filed for 4 patents and presented my work to a global audience at ReMars, Vegas 2019.

Led machine learning engineer on the python/tornado/akka/scala/spark/sql/redis stack. Responsibilities included research, development and iterative improvement of the Data Science pipeline. Selected projects ....

• Model Evaluation : Ideated and implemented the core model evaluation framework. Techniques included stratified CV and AUROCs.
• Automated Model Generation : Setup automated systems using S3 for feature engineering and continuous model creation and periodic deployment. Efforts resulted in classification accuracy improvements of 7% into the last decile
• Successfully filed for 1 patent and presented at the Detect 2017, Washington DC.

Continued with the same team as iSocket on larger out facing projects with added responsibility. Selected projects.....

• Recommendation Engine : Ideated and engineered a recommender system using collaborative filtering to connect advertisers to the long tail(comscore>1000) of publishers.
• Creative Management : Led the engineering implementation of the full automation of the upload and synchronization of ad-placements across DFP and Rubicon systems.

Designed, developed and implemented features on the Node.js/PostgreSql/MySql/RabbitMq/Redis/ElasticSearch stack. Was an integral part of a small (Selected Projects include....

• Order-Change Feature : Was part of a 2 member team working across the stack that engineered the crucial cross-platform feature which allowed buyers and sellers to negotiate back and forth on orders.

Developed, maintained, and iteratively upgraded cutting-edge Real Estate web software on the Linux/Ruby/Rails/PostgresSQL/Heroku stack. Designed and implemented product, collaborating with project managers, engineering team, and client representatives. Provided technical leadership and managed collaboration for engineering team. Selected achievements include ....

• Inventory Aggregator : Designed and implemented aggregator in Ruby to populate and normalize data from 60+ unique nationwide sources of real estate inventory making RoomHunt the 2nd biggest source of inventory in major cities.