I am a Machine Learning, Deep Learning and Computer Vision Engineer with a background in Computer Science and Software Development. I have 2+ years of professional experience in developing software and applying AI methods to real-world challenges across different domains, including computer vision and network security. I am passionate about building computer vision systems that can perform visual tasks efficiently. I enjoy keeping up with the recent advancements in the field and continue to grow my expertise in it. If you are looking for a Deep Learning or Machine Learning Engineer, please check out my CV, LinkedIn, GitHub or email me.
- Programming: Python (numpy, pandas, sklearn etc.), C, Software Architechture and Testing
- DL: DL Architectures, Computer Vision, OpenCV, Keras, PyTorch, Tensorflow
- ML: ML Methods and Theory; Developing, Deploying and Debugging ML Pipelines
- Back-end: Django, Flask, REST API, Database Design, SQL, NoSQL, MongoDB, PostgresSQL
- DevOps: Docker, AWS, Google Cloud
Product Detection and Classification using Deep Learning
During my deep learning internship at Focal Systems I worked on the tasks of detection and classification of products on grocery store shelves. One of my tasks at the company was to prepare this demo for a conference:
Tech stack: Python, Keras, Docker.
Joint effort with Focal Systems's team.
Can Neural Networks learn L1-Norm Operation?
As a part of a coding challenge I was asked to create and train a Neural Network to take a random array of real valued numbers that is variable in length, and let it learn the L1-Norm of the array.
Tech stack: IPython, Tensorflow.
A term project for the course CS-E4820 - Machine Learning: Advanced Probabilistic Methods at Aalto University, taught in spring 2017. The goal of the project was to construct a Bayesian regression model for incomplete censored data. We considered polynomial regression and used the EM algorithm to obtain a Maximum a Posteriori (MAP) estimate for the parameters of the model. We used Bayesian Information Criterion (BIC) to select the degree of the polynomial.
Tech stack: Matlab.
A term project for the course CS-E4600 - Algorithmic Methods of Data Mining at Aalto University taught in fall 2016. The goal of the project was to compute statistics - mean, median, diameter and effective diameter - for large networks. We implemented the exact algorithm (only feasible on small networks) and approximate algorithms: sample random pairs, sample random sources, Approximate Neighborhood Function (ANF). To speed up the running time we have parallelized the algorithms.
Tech stack: Python, SciPy, Cython.
Joint effort with Max Reuter.
Carsus is a python package for storing and manipulating atomic data, such as atomic masses, ionization energies, levels and transitions. This data is used by TARDIS - an open-source scientific code for rapid spectral modelling of supernovae. Carsus downloads and parses data from a number of sources, stores it in an SQLite database and outputs it in the HDF5 format.
Tech stack: Python, Python scientific stack, pandas, SQLAlchemy and SQLite, pyparsing, pytest.