Treffit.fi -- Acquired by City Digital Group:
End-to-End Machine learning pipeline for Automated image processing + entity extraction based on
Resnet50 Neural network architecture
End-to-End Machine learning pipeline for semantic similarity based automated spam banning system based on auto-encoders and, Locality Sensitive Hashing for efficient retrieval at scale
[Python, Numpy, Tensorflow, AWS Lambda]
Aller miscellaneous/other:
- Dating related, recommendation system
- MariaDB -> AWS parquet based pipeline to leverage Athena Columnar database for analytics
- MSSQL to snowflake transpiler in order to fully migrate MSSQL based internal services into snowflake
- Python pandas based wrapper to guarantee a lower expense in snowflake, by caching existing results.
- Docker based airflow deployments and automatic pipeline-creation from custom files in s3
- Thorough training of 4 different staff members in topics ranging from python to machine learning, snowflake and, airflow

-- Acquired by Epic Games
Co-lead the preparations and implementations for the Kamu Junction challenge 2017 on Europe's Largest Hackathon: https://hackjunction.com/challenges/play-beautiful [https://junction2017.devpost.com/]

As part of the core business, work to aid in the detection of cheaters & binary data corruption machine learning models Included:

Recurrent Neural Network LSTM based multiclass classification for repeated user fingerprinting.
Gradient boosted trees based Aimbot Detection,
Recurrent Convolutional Autoencoder based Similarity-based information retrieval engine for binary-data similarity.

In order to deal with anomalous events in our live user data:
Kafka streaming data anomaly detection engine algorithms included:
- Convex Hull based anomaly detection
- Bayesian Changepoint shift detection
- Kolmogorov smirnof test among others
To deal with large scale data -> approximate & streaming algorithm based online learning:
- Streaming statistics generation at O(1) space complexity
- Streaming clustering algorithms