Experience

Metaphysic.ai

• United Kingdom
• Technology, Information and Internet
• 1 - 100 Employee

• Senior Generative AI Researcher

  • Mar 2022 - Present

  Developing temporally consistent AI methods for the generation of hyperreal synthetic media in the film industry, working closely with VFX professionals. Developing temporally consistent AI methods for the generation of hyperreal synthetic media in the film industry, working closely with VFX professionals.



Chess Mate

• Bucharest, Romania

• Founder

  • Nov 2021 - Jul 2022

  Lead technical and business development for a cross-platform mobile app targeted at professional chess players, using SOTA proprietary computer vision algorithms, in a team of 3. Conducted market research with forms and interviews with professional chess players, pivoted from original idea, improved business strategy and held multiple pitches in front large audiences with live broadcasting, during Innovation Labs, the largest startup accelerator program in Romania; Built on top of my earlier work to develop a computer vision algorithm for real-time parsing of chess moves from videos taken with the mobile app. It works on most standard chessboards, by self-calibrating on each individual chessboard and continuously learning throughout the game. Show less



METAMINDS

• Romania
• Information Technology & Services
• 1 - 100 Employee

• Machine Learning Engineer

  • Jul 2020 - Jul 2021

  Early member of an AI startup where we developed a rep counting and posture-correcting fitness app from scratch. My focus was on developing and deploying a lightweight AI pipeline that processes videos in near-real time. I researched deep learning tasks & architectures (object detection and tracking, object segmentation with knowledge distillation, keypoint detection, and classification), gaining my first mentorship/management experience. I've also collaborated directly with the mobile development team to continuously integrate our research into the app. I worked on deploying the pipeline on a local server, using Flask and Cassandra. I scaled the deployment server with distributed queues and adapted it to use AWS services. Finally, I scaled the pipeline further with a serverless service. I developed a web application with CRUD operations over cloud databases and buckets, for semi-automatic video annotation by using self-supervised learning, computer vision and signal processing (Fourier Transform, Convolution, Autocorrelation). I built multiple computers for machine learning, including the initial server. Conducted technical interviews. Show less



Sparktech Software

• IT Services and IT Consulting
• 1 - 100 Employee

• Machine Learning Engineer

  • Jul 2019 - Jun 2020

  Researched a date OCR algorithm with ~90% accuracy (Tesseract OCRhad 70%) , trained on ~800 images and artificial data, where I gained hands-on experience with GANs, autoencoders, EfficientNet CNNs, GRU & LSTM RNNs, transfer and multi-task learning; Implemented new features in a large Python codebase; Researched anddeveloped solutions for a data quality assurance and augmentationproject; Held a demo on Google Cloud AutoML; Researched time seriesforecasting algorithms and frameworks on COVID-19 data; Brieflyresearched recommender systems. Show less

• Machine Learning Research Scholarship

  • Feb 2019 - Jul 2019

  Developed an AI pipeline for parsing chessboard states from single photos, composed of an algorithm for chessboard detection with 58% accuracy (SOTA had 24%) and one for chess piece recognition with 93% accuracy (SOTA had 87%), which used deep learning (image segmentation CNN, multi-task MobileNetV2), computer vision and traditional machine learning; Created large datasets (including images generated with video games), mostly annotated automatically; Used Google Cloud for training.