Fadi Bakoura

I find meaning in life by helping to reduce people's suffering and increase their quality of life. I'm achieving this by thriving to build and collaborate to produce high-quality (scalable, extendable, reliable, available, reusable, secure) software products that have the biggest impact.

I'm developing bioinformatics solutions to help scientists in their drug-discovery workflows. I have also strong background in Machine Learning (developed AI solution for a large telecom company). I collaborate and communicate well with team members and stakeholders. I follow the agile development process (experience in prioritizing tasks, time management, short-term planning, take rapid feedback from users and stakeholders). I strongly adhere to software engineering principles and practices (clean code - OOP design - TTD - SOLID principles - design patterns).

• Visualization: DNA Vector Editor - Antibody Numbering - Restriction Enzymes - Peptide Cutter - Sequence Alignments.
• Workflows and tools: DNA Cloning - Antibody Engineering - Protein Analysis - Sequence Alignments - BLAST Search -  Sequence Annotation & Prediction.
• Data: integrated Drugs, Proteins, and DNA from UniProt, ChEMBL, PDB, and more. (I'm responsible for those data migration to the Genophore's evolving Database).
• I used AWS Lambda function (with SAM deployment) for periodically integrating about 80 G.B data from a third-party service, achieving processing speed (compared to regular data pipeline) from about 1 day to 1 hour. 

I helped in the analysis, design, and I worked on the implementation of predicting subscribers' likelihood of leaving the company. A brief description of the system: on a daily basis, the system predicts new churn scores for 6 million subscribers. It fetches the data from the internal warehouse database, Then, it feeds a large 3D array to the deep learning model. A highly optimized operations are developed in order for the pipeline to run in an acceptable time. In fact, all steps (comprised of data fetching, transformation, prediction, evaluating, and publishing the results) take approximately four hours. The model training is also automated.

• Spoofing and countermeasures for Automatic Speaker Verification using Gaussian Mixture Models with I-vectors.
• Dialect identification using SVM, Random Forest with Sickit-learn library.
• Restoring Text Diacritics (End-to-end deep learning).
• Swarm Robots, Real-Time Adaptive Shape Formation, shapes were fully determined by the user by drawing using Unity3D.
• Interactive virtual reality game developed with google cardboard SDK and Unity3D.
• JavaScript compiler using JAVACC library.