Professional Projects

[Python, PyTorch, Numpy, Pandas, GeoPandas, Scikit-learn, XGBoost, FBProphet, NLTK, SpaCy, Gensim, NetworkX]

​

• Machine Learning Engineering (NLP / Computer vision):

  • Built an image annotation pipeline with CLIP model as backbone and Faster/Mask R-CNN as object detector/segmentator;

  • Created a duplicate image detection dash app with imagehash and TSNE;

  • Developed NLP-based models and  pipeline to match and group similar products offered by different merchants;

  • Wrote bespoke Named-Entity Recognition library to identify product features;

​

• Time-series forecasting using a wide range of ML techniques including Lasso regression, SVM, random forests and Gradient Boosting methods with feature engineering.

 

Personal Projects

[PyTorch, Tensorflow, Transformers, OpenCV, Pillow, Scikit-image, Albumentations]

​

Conditional Text Generation given title and keywords by fine tuning GPT-2: Medium Article, Colab notebook.

​

Twitter sentiment analysis on US Election 2020:

• Article on Medium;

• Corresponding Colab notebook on fine-tuning RoBERTa on TPUs.

​

Prove of concept and feasibility for a prospective client:

• Performed classification and sentiment analysis by fine-tuning RoBERTa on Google TPUs:

  • Colab notebook on classification;

  • Colab notebook on sentiment analysis.

​

Data Visualization

[Matplotlib, Seaborn, Dash, Plotly, Kepler.gl, Yellowbrick]

​

• Plotly Web-App in showing England and Wales average house prices by postcode sector: https://ukhouseprice.project-ds.net/

• A Kaggle notebook Taxi Trips as Air Distance – Animation.

​

Algo-Trading

[Matlab, Julia, Java]

​

• Developed a java-based proprietary algo-trading system to trade on Interactive Broker, with bespoke models developed and back-tested in Python and Julia.

• High-frequency algo trading in spot FX on EBS and Reuters with an average daily volume of EUR 6 billion. Proprietary models developed in Matlab.

​

​

Kaggle

​

Competition contributor:

• Top 10% in Quora Question Pairs NLP competition;

• Top 14% in Data Science Bowl 2017 on lung cancer detection from medical images;

• Top 12% in Amazon from Space satellite data image classification competition.

​

​

​