Jan 2022 – Jan 2025 · 3 yrs · Greater Milan Metropolitan Area

• Made the biggest AI student club in Italy, 3rd in Europe by size.
• Connected thousands of students to research opportunities, $100,000+ raised, several papers published out of it etc.

Jan 2021 – Jan 2022 · 1 yr · Sofia, Bulgaria

• Ran a 21K mountain-climbing race and filmed it to raise awareness about skin diseases. Generated 100,000+ views, $1,000+ in donations, interviewed on national TV streamed to 6M people, etc.

Jan 2020 – Present · 6 yrs 3 mos · Tel Aviv District, Israel

• Worked on cognitively-inspired VLM architectures. Designed a SOTA vision-language architecture that uses 10x less compute. Supervisors: Shimon Ullman and Daniel Harari (both from MIT).
• Published a few papers in CVPR, ECCV etc. including a Spotlight.

Jul 2019 – Aug 2019 · 1 mo · Rehovot Area, Israel

• As part of my participation in the International Summer Science Institute (ISSI) program, worked on genome analysis algorithms based on BLAST. Throughout the duration of my stay, I managed to gather proof suggesting that soil bacteria is adapting to conventionally used soil fertilizers and summarized the results in a scientific report. Main technologies used for the project were Python and MATLAB.

Jan 2020 – Jan 2021 · 1 yr · London, United Kingdom

• Led research and development efforts to entirly implement the following systems as part of the company's workflow:
• Pipeline for OCR of medical documents. Technical stack: image enhancement,
• machine learning, NLP, Azure, SQL
• Telehealth system for transcribing patient-doctor consultations. Technical stack: AWS, Python, SQL
• Wearable device big-data collection & aggregation. Technical stack: machine learning, Withings API, multi-threading
• Several education workshops: How2Code Python, Computer Vision Basics
• In addition, supported customers on Live environments & kept active communication with multiple technical support teams from external organizations - which outsource some of our services (wearable device hardware, website front-end development ect.).

Jan 2019 – Present · 7 yrs 3 mos · London, United Kingdom

• Selected as 1 of 500 applicants annually to be a member of the Livery. They're an exclusive network of scientists with a mentorship component. My mentor was the guy who invented MRI scanners (Dr. Barrie Reece).

Sep 2018 – Nov 2019 · 1 yr 2 mos · Bulgaria

• Worked on two computational biology projects with applications in medical image analysis. Both projects were entirely implemented in Python and MATLAB. Project descriptions:
• 1. Novel method for skeletal age estimation based on cranial suture analysis
• This project’s aim is to objectify and improve the accuracy of age estimation methods based on cranial suture analysis by elaborating an algorithm for automatic assessment of the suture closure degree in cross-section and assessing its relation to aging. For the purpose, volumetric images of dry skulls, generated by an industrial micro-CT system, are used. The images are analyzed by VGG-16, a semantic segmentation neural network, and the C-Means clustering algorithm.
• A number of quantifiable metrics about the degree of suture fusion are recorded. Such metrics have been collected for a large sample and correlated with the AAD (age-at-death) using linear regression. The produced model manages to reduce the error of predicted age estimates on validation samples over 3 times compared to the conventionally used methods. Project article:
• https://www.sciencenewsforstudents.org/blog/eureka-lab/ISEF-2019-new-forensic-technique-better-gauge-age-death
• 2. Super-resolution on medical images
• Low image quality is a common problem when dealing with CT data taken in vivo. The radiation exposure is the determining factor for quality, and as biological tissue is very vulnerable in that sense, commonly, little radiation is used. As a result, the fine details on such tomographical data is usually lost or hardly visible. This project aims at improving the readability of selected regions on medical images, by implementing a series of image processing techniques. Those include super-resolution with bi-cubic interpolation, kernel-based deblurring and SIM (Sphere Intersection Method) denoising. The algorithm managed to increase the NIQUE (Naturalness Image Quality Evaluator) score of medical samples over 2 times compared to the original.

Sep 2016 – Oct 2016 · 1 mo

• Participated in a selective programming academy organized by SAP Labs Bulgaria. Throughout the program, explored a variety of programming languages and completed an individual project in Java at the end of the program. The project was an Android game, widely known in the real world as the "Shotgun" game. It's simillar to "Rock, Paper, Scissors" in that it has three main moves, but is a lot more strategic and not luck based.
• The mobile app was presented at the National IT Olympiad and a national competition for mobile applications AppsNPMG. It managed to win first place for the age category at both competitions. The implementation included developing a custom client-server framework in Java, which utilized the TCP protocol and was also programmable by custom plugins. The mobile application was entirely created in Java, using Android Studio.

Jun 2016 – Dec 2016 · 6 mos · Greater Boston Area

• Attended the Wolfram Summer School for student researchers. Throughout my stay, created a decoder for human-generated morse code sound. The implementation involved work with clustering algorithms and frequency thresholding. Entirely implemented in the Wolfram Language.
• Thereafter, was selected to partake in the Wolfram Mentorship program. There, I entirely implemented a classifier for animal tracks & was able to successfully distinguish between the four most commonly encountered animal families - canines, felines, rodents and ungulates.

Nov 2014 – Mar 2015 · 4 mos

• Worked on a psychology research project with Dr. Ivan Bardov, an academician in Experimental and Developmental Psychology. Developed a custom software implementation of the Max Luscher personality test, which was later on administered on a sample with research purposes. The development was carried out in C# and Java.
• The test displays various geometrical shapes and allows the participant to color each one by choice. Based on several repetitions of the tasks, an overall personality analysis is created.

Sep 2014 – Jun 2015 · 9 mos

• Took part in an international FLL robotics team. Our page:
• https://www.facebook.com/KASAPIN40/
• Throughout the year, we build and programmed a robot to pass a number of preset challenges. Those included following a line, scoring a goal at soccer and moving/lifting various objects around a field.
• Besides building a robot, the FLL competition format includes a research project on a different theme every year. For our participation, we had to create a research project on the topic of "World Education". We made an Android application for studying foreign languages. Our team won 2nd place in "Mechanical Design" at the World Championships in Johannesburg. We won 1st place at two other international competitions - in Bulgaria and Greece.