Research

Here, I have showcased some of my interesting projects:

Traffic waves

Traffic-Waves is a voluntary project focused on daily traffic predictions in Paris, utilizing data from Open Data Paris. The project leverages deep learning techniques to analyze historical traffic data and make predictions for daily traffic patterns in Paris. This aids in providing insights for commuters and city planners alike. The system is deployed on AWS, utilizing various services to process and analyze the traffic data efficiently.

pneuma treatment

I developed an automated way to treat the noise and anomalies (in the form of unrealistic peaks) in the acceleration values of a number of vehicles in the pneuma dataset. The algorithm uses a combination of low-pass filters (Savitzky-Golay filter, Gaussian filter) to remove the noise. A machine learning model (XGBoost) is used to reconstruct acceleration time-series and detect anomalies. If anomalies are detected, they are removed. The outputs are the processed time series profiles of vehicle speeds and accelerations without high-frequency noise and unrealistic acceleration peaks.

actrys

actrys is a Python-based platform designed for calibrating traffic simulations in SUMO (Simulation of Urban Mobility). The platform follows a step-wise approach for sequential calibration of mesoscopic simulations. It uses heuristics, ensembling, and Bayesian optimization to efficiently achieve its goals.

TraMPA

In TraMPA, we investigated the use of openly available data for the development of transportation models. The proposed project investigated the ever-growing amount of data that can be obtained from a diverse number of open data sources available. The analyses were conducted on different levels of open-source data availability to perform sensitivity analyses and meta-model analyses for the investigation of the model performance based on real-world scenarios.

Scopus caller

I developed a tool called scopus-caller to quickly collect the meta-data of thousands of scientific articles from the official SCOPUS database. I packaged the tool as a Python pip package and also hosted the tool as a streamlit app called Scholar scout 🐦‍⬛. This tool is helpful for researchers and practitioners to quickly scan through millions of articles for relevant texts and speed up their literature discovery and review process.

Magnet

We developed a recommendation system named Magnet using NLP methods to assist beginner and intermediate users of AF3, an IDE for Model-Driven Development, in learning the tool. In this project, I led the data processing, machine learning (ML) model training, and the integration of the ML pipeline with the AF3 tool. For further details, you can view our demo video and publication.