profile image

Luc Atakpa

Software / Machine Learning Engineer

I am open to freelancing or full time remote opportunities

My Apps

About Me

Hey 👋🏾 I'm Luc, a software and machine learning engineer. I have hands-on experience building backend applications and machine learning / deep learning models for various use cases in Speech Recognition, Natural Language Processing, and Computer Vision. I value efficiency when writing programs, and applying best practices to ensure high code quality.
I enjoy writing computer programs, building machine learning models from scratch to deepen my understanding of the underlying algorithms, and also applying them with high-level libraries to real-world data. I like sharpening my skills and working in friendly environments, on interesting/impactful projects.

Work Experience

Research Associate - CyLab-Africa (2022 - Present)

Created programming materials for a "Probabilistic Graphical Models" course and expected to work on machine learning projects at CyLab-Africa.

Data scientist (Part time) - VIDA (2022)

Explored new geospatial datasets to assess the insights that can be extracted from them and the value that they could add to the company's customers.

Data Science Intern - VIDA (2021)

Wrote processing scripts for geospatial data with Geopandas. Converted Jupyter notebooks into command line interface scripts, and used ClearML to log and track experiments. Made database change management easier by creating and maintaining database schema migrations with alembic.

Software Engineer - MIFY (2019 - 2020)

Built a proof of concept for a speech recognition system for a local language in Benin named Fon, and deployed it on Google Cloud Platform. I also maintained the company's website.

Software Developer Intern - Laboratoire de Recherche en Sciences Informatique et Applications (2018)

Added a set of features to a grading system for an online test platform for students. The system is based on the open source project INGInious.

Some Projects

Web Mapping Application Web App

Developed a web mapping application for a surveyor's company. It allows them to visualize spatial data in a Web browser, query the data and get building footprints' attributes on the interactive map.

Tools: Express.js, PostGIS, Geoserver, Leaflet.js

Code Draft

Extreme Multi-Domain, Multi-Task Learning With Transformers Research Paper

Researched “Extreme Multi-Domain, Multi-Task Learning With Unified Text-to-Text Transfer Transformers” with a team. The main goal of the research was to assess how the multi-task learning paradigm with the unified text-to-text transfer transformer (T5) extends beyond natural language tasks into extremely different domains like playing chess and generating code, and how to better adapt it to multi-domain learning.

Check the Paper

Speech Recognition for Fon Web App

Built a proof of concept for a speech recognition system for a local language in Benin named Fon, and deployed it on Google Cloud Platform. The system was developed using the open source tools Kaldi, Pykaldi and a public dataset available here.

Code Draft

My GitHub