Norbert Kenderesi Email & Phone Number

Middletown, Nj, Us

Middletown, Nj, Us

New York, Ny, Us

Budapest, Pest, Hu

Electric meter reading machine learning models and application - Designed, implemented and productionized the models (F-CNN and CNN) and the application. - Optimized model for mobile and CPU usage. On a Pixel 3 digit detection is real time (~130ms) meter reading detection with high accuracy is in 2-3 secs in most cases. On a 4 core desktop CPU reading is under 1 sec on most images. - Model supports 60+ different hardware models with both digital and analog displays. - Overall accuracy of meter readings is ~75%, with some hardware being ~93%. Digit recognition is ~93%, hardware type recognition is ~85% (some hardwares are very similar). - Trained with data gathered from 2500 images and tested on a 4K validation set.- Kicking off training required generating the training data and the labels by models derived from simple digit recognition, trained with open source datasetsLead and overseen the development of several outsourced projects for both Android and iOS (lead 4, overseen 5+).

South San Francisco, California, Us

Mountain View, Ca, Us

Modernize Waze Logging infrastructure. - Researched feasibility, proposed, designed and leading moving application logs to Google Cloud Logging.- Run pilot deployment and planned staging deploymentMove Waze logging for flogger. - Researched feasibility (run benchmarks, figured out missing pieces, proposed fixes) and wrote proposal. - Implemented flogger changes. - Researched and leading implementation of automatic code refactoring tool based on error-prone.Making Youtube Music intelligently refresh Home Page recommendations- Designed a whole system to make decisions on when to refresh recommendations possible- Implemented and drove implementation- Designed and run experiments to track usefulness and drive launch decision of features as well as decide the best settings.- Worked on the whole stack from server side data gathering to passing information on to the mobile apps and implementing some features in the Android app as well.Engineered data pipelines and prepared data for intelligence projects on GSuite.- Designed and optimized huge map reduce jobs running daily and generating couple hundred terrabytes of data daily- Designed and built tooling to make usage of said pipelines easyWorked on Android and iOS Google Docs. Some of my contributions include- Co-designed generic - platform and language agnostic - rendering api used on mobile for Slides, Sheets and Docs- Rendering performance analysis and optimizations- Made context menu accessible on Android, designing a solution which worked on older Android version as well.

Lancing, Gb

LaithWaite's Wine iPad application development

Zurich, Ch

Working on enhancement of a huge, heavily multithreaded, Java, Spring and Oracle based system (FBI project).

Newtown, Pa, Us

- Memory optimization and database refactoring of an iPad news application developed for News Corporation. The peak memory usage went down from 25MB to 10MB – sadly the project has been shut down 85% ready.- Was secondary lead on the same project; using scrum with 2-week sprints.- Lead development and successful launch of middleware system between Apple and O2 systems for the UK iPhone introduction. The system was able to serve 1500+ activation/minutes during peak.- Acting as scrum master for the Hungarian part of that team; 2-week sprints