Robert Steen Email & Phone Number

Redmond, Washington, Us

Office Chat Copilot - Co-Inventor and co-architect of Office Chat CopilotPersonalization of ML - Driving innovations in company-wide infrastructure in order to deliver v1 personalized ML - Applying user specific knowledge to personalized proactive and search based featuresDelivering intelligence to Office experience - Partnering with product teams to deliver intelligent Office search scenarios by identifying the correct problem, data, and approach in order to train and deliver ML models which improve a user's Office experience.Expertise in NLP & Transformers

Redmond, Washington, Us

Full stack applied ML researcher focused on delivering video understanding scenarios for Office. Perform ML research, data collection, model training, measurement, service architecture, service deployment, and service integration.Auto video chaptering• Drove project to automatically segment videos and title the segments, making the videos easier to consume. Defined problem, managed data collection and annotation, and created evaluation framework. Trained segmentation models (based on BERT, DNN, LSTM, CNN, and more) and partnered with researchers who trained additional models for both segmentation and segment titling.Auto video trim• Architected and implemented services which detect the start and end of the content of a video recording for Stream in order to remove this “waste”.Object tracking• Partnered with researchers to deliver an object tracking model for a video editor, enabling a user to click on a single point of an object and add a predefined effect (eg. Fireball) such that the effect tracked to that object across video frames.

Redmond, Washington, Us

Smart Headphones• Prototyped a head gesture recognizer, orientation specific car directions, and orientation aware landmark identification by soldering sensors to a microcontroller, attaching to headphones, and writing microcontroller code to detect head gestures and orientation, influencing our future hardware“Growing Up with Windows” Pitch• Pitched a set of investments across the company to enable children as young as 2 years old to use Microsoft products and services by collaborating with teammates, investigating the market, and prototyping physical object interactions and a new shell for Windows that suits their abilitiesML applied to Boggle• Built “Virtual Boggle” game by training a custom OCR for Boggle dice using AzureML for a 2-day Hackathon project that attained OCR with ~98% accuracyAR + Physical Objects• Rendered an interactive Holographic UI that tracked with the Surface Dial on the HoloLens by using image recognition to detect a Surface Dial Shell Chatbot• Investigated natural language interactions to control remote computers by using socket programming, Microsoft AI bot framework, LUIS, Twilio, and system APIs

Redmond, Washington, Us

Email Voice Agent & Voice Agent Architecture• Launched a new framework for multi-turn voice agents powered by the cloud, by driving integration with cloud and framework teams while working on the first voice agent (Email) to use the framework, enabling real-time updates to the voice agents, quicker development, and simpler translation to new languages

Redmond, Washington, Us

Text Rendering APIs• Automated support for new scripts (languages) by implementing an algorithm based on published Unicode files, reducing the amount of work for each new script from weeks to seconds• Improved text layout for bidirectional text by designing and driving a company-wide improvement to the Unicode bidirectional text algorithm, which is now part of the Unicode standard.Touch Text Selection• Delivered touch text selection in Windows 8 by implementing an interaction model in Internet Explorer and web appsWindows JavaScript Platform• Facilitated a new development platform by designing and implementing AppBar, Menu, and Settings controls for JavaScript Windows apps, making it easier to develop apps

Redmond, Washington, Us

Developed UI Regeneration Tool, providing real-time viewing of localized UI for translators. Allowed translators to find localization bugs while in the process of localizing a UI.

San Jose, Ca, Us

Managed cross-functional team communications.Setup lab test bed including virtual machines for simulating real-world network.Developed scripts to automate testing process of MeetingPlace through simulation of users, which included user actions, parsing files for information, and timing analysis.