Tech lead of an applied research team focused on video understanding. Led development of computer vision projects ensuring product-level quality. Implemented highly efficient software in modern C++ and Python from end to end: from video streaming and processing to delivering customer-specific BI/alerts, from implementing multi-thread computer vision algorithms to scalable cloud services. Research in computer vision and machine learning.- Led team development with best practices and tools: C++ Core Guidelines, unit tests, integration tests, code reviews, code style, security & privacy reviews. Cross-team efforts to establish unified development practices.- Designed and developed a high performance pipeline for near real-time video understanding (object detection and classification), coordinating traditional computer vision algorithms (image enhancement, background subtraction, object tracking) and CNNs (Faster R-CNN, ResNet) with live streaming and business logic.- Continuous evaluation and improvement of the pipeline. On CPU, average utilization below 1 core per camera stream, below $1 per camera per day on Azure retail price.- Research: Development of CNNs for automatic camera calibration and lens distortion rectification from single image.- Research: Developed out-of-distribution detection methods using Batch Normalization features and SVM with applications to anomaly detection, camera malfunction detection and dataset selection.- Designed and developed cloud and edge production services deployed on 4 customers supporting over 350 simultaneous camera streams - Service Fabric, IoT Edge.- Experience with the full machine learning lifecycle: data acquisition, annotation, training, evaluation and deployment.- Customer interaction on different application verticals: workplace safety, public security, traffic monitoring, hospital safety.- Collaborations on 1P and 3P projects: PoCs, collaboration with Microsoft Consulting Services, Azure Computer Vision, etc.

Member of the Industrial Cloud Technologies group with focus on cloud service research and stream computing.- Created a novel dynamic resource allocation framework for cloud systems and for auto-scaling of stream computing applications. Methods predict future system load based on application-specific metadata.- Part of the IBM Watson bootstrap initiative in Latin America, presenting in events and universities.

Python developer in an e-commerce monitoring start-up. Customers included all major online retailers in Brazil.- Developed cloud-based Python web crawlers and scrapers for monitoring product prices, availability and delivery. Led integration with other services (like Google AdWords) and ecommerce platform APIs. Also worked on item duplicity detection and customer-specific BI.- Developed a dynamic price update framework based on supply/demand monitoring and microeconomic models.

Member of the Social Team. Spent first year working on the Orkut social network, later on Google+ internal systems for user data management and abuse. Developed large scale cloud systems in Java.- Built Orkut’s web REST API and public Java SDK for 3rd-party integrators.- Built Orkut’s Android app. Over 5M installs.- Developed Google+ user data administration service: a central service mediating all employee access to user data, ensuring privacy, security and auditability. Grew service to other products integrating with 40+ other product backends.- Developed and maintained high availability geo-distributed cloud systems handling massive amounts of data (Bigtable, Borg, MapReduce, gRPC, Protocol Buffers, Guice).- Participated in large development projects interacting with multiple teams across different countries.

Member of a university research lab specialized in voice over IP. Undergrad part-time intern until 2008, full-time onward. Focus on quality monitoring of multimedia networks. Development leader.- Research, development and deployment of voice quality models and tools, including integration with VoIP software, like Asterisk. Focus on ITU-T recommendations.- Created a C++ call emulator supporting different codecs, quality models and stochastic network impairment models (packet loss, latency, jitter, etc).- Developed Asterisk module to integrate a proprietary videoconferencing system (audio, video, text, etc) to standard VoIP protocols (SIP, H.323) and PSTN.- Led development of fone@RNP, a VoIP service connecting over 100 federal institutions in Brazil at the time. 100% made of open source software, the system was developed as a Debian-based distribution. Developed system management web tools and taught classes to site admins. C, Perl, Bash, Javascript, PHP.