Deepak John Reji
Senior Data Scientist at Carelon Global Solutions Ireland
Limerick, Ireland
Actions
Deepak John Reji is an experienced NLP practitioner and researcher who specializes in developing and designing solutions for data science products. He is passionate about domain-centric data science development and advocates for responsible and ethical AI development. Deepak enjoys working with social science and environmental data and creates videos showcasing various prototypes and AI tutorials. Additionally, he hosts the D4 Data Channel podcast, where he interviews industry experts and academicians in the fields of Data Science and Sustainability.
Deepak strongly believes in the power of open source and the democratization of AI. He actively contributes to open-source projects and shares his models and packages in Huggingface and PyPI repositories. His recent research focuses on Immigration and Integration Research, as well as studying human and community interactions in third places using AI and data.
Linkedin: https://www.linkedin.com/in/deepak-john-reji/
Youtube: https://www.youtube.com/@deepakjohnreji
Google Scholar: https://scholar.google.com/citations?user=4azuuXAAAAAJ&hl=en
Area of Expertise
Topics
Dynamic Analysis of Immigration Attitudes: Event-Driven Modelling with NLP & Temporal Networks
How do major events cause seismic shifts in public opinion? This talk explains a data science workflow built in Python to model how immigration attitudes in Ireland are reshaped by critical events rather than evolving gradually as classic sociological theories predict.
In this talk, I will walk through the process of using Python to trace these shifts, drawing from years of Irish news stories, tweets, and national surveys also explaining the NLP tooling (Gen AI, transformers), dynamic network analysis, and agent-based social simulations to uncover short-lived, emotionally driven spikes in discourse after high-profile events (eg: Cologne incident in Germany). Alongside computational techniques, attendees will see how temporal network analysis isolates the “signal” of event-driven changes using custom exponential decay kernels, and how attitude-network maps reveal symbolic connections in media framing.
Key technical components include:
- Data engineering pipelines for scraping, cleaning, and normalizing multi-source text corpora
- Advanced NLP: topic modelling, sentiment trajectory tracking, gendered discourse with transformer models
- Temporal graph modelling: visualizing the spread and clustering of attitudes post-event with ResIN network model, and custom weighting functions
- Agent-based simulations for hypothesis testing: comparing linear (modernization) vs. reactive (event-driven) models
The PyCon community will benefit from this talk, which explains how Python empowers interdisciplinary research, bridging social science theory, real-world AI, and robust engineering. It also gives an understanding to viewers to scale up NLP pipelines or build temporal network models, and leaves with patterns and architectures applicable to a range of challenges, and a fresh perspective on using Python to interrogate complex societal phenomena.
Why is this relevant?
With current political and social debate often hinging on short bursts of public concern, understanding how attitudes are formed and how to model them with Python has broad relevance for data scientists, AI practitioners, policy-makers, and anyone working on real-world text or network problems. The session prioritises transparent code, reproducibility, and technical detail, making it suitable for both academic and industry audiences.
Dbias - Detecting Bias and ensuring Fairness in AI solutions
Recommender systems, information retrieval, and other information access systems present unique challenges for examining and applying concepts of fairness and bias mitigation in unstructured text. This paper introduces Dbias, which is a Python package to ensure fairness in news articles. Dbias is a trained Machine Learning (ML) pipeline that can take a text (e.g., a paragraph or news story) and detects if the text is biased or not. It detects the biased words in the text, masks them, and recommends a set of sentences with new words that are bias-free or at least less biased. We incorporate the elements of data science best practices to ensure that this pipeline is reproducible and usable. We show in experiments that this pipeline can be effective for mitigating biases and outperforms the common neural network architectures in ensuring fairness in news articles.
1H EnvBert: An NLP model for Environmental Due Diligence data classification
EnvBert is a Natural Language Processing (NLP) package built on top of DistilBERT, focused on Environmental Due Diligence (EDD). It comprises four functions: text classification, relevancy detection, ranking and fine-tuning. The environmental dataset was used for fine-tuning DistilBERT’s performance to develop the EDD model. It is hosted as an inference Application Programming Interface (API) on Hugging Face Hub. The EDD model with custom vector representation constitutes EnvBert, it can be installed using the package manager pip.
Pycon Ireland 2025 Sessionize Event
Pycon Ireland 2024 Sessionize Event
Pycon Ireland 2023 Sessionize Event
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top