About Me
Hi there, I am Kelechi, a Ph.D. student at the University of Colorado Boulder, under the guidance of Prof. Sidney D'Mello. My research focuses on the intersection of educational Natural Language Processing (NLP) and Human AI Interactions, where I use machine learning techniques to analyze user interactions and speech data to uncover cognitive and behavioral patterns in these environments. I aim to enhance the understanding of how both learners and educators interact in these contexts.
I hold a Bachelor of Science in Computer Science and Statistics from the University of Nigeria, Nsukka, and an Associate Teaching Degree in Computer Science and Mathematics Education from the Federal College of Education, Kontagora. Before pursuing my doctoral studies, I worked as a Software Engineer in Nigeria, developing practical technology solutions. This experience informs my research, which seeks to bridge technology and education to improve learning outcomes.
Recent News
Publications & Presentations
Selected Publications
'It feels like we're not meeting the criteria': Examining and Mitigating the Cascading Effects of Bias in Automatic Speech Recognition in Automatic Feedback Interfaces
Kelechi Ezema, Chelsea Chandler, Rosy Southwell, Niranjan Cholendiran and Sidney D'Mello
Conference on Human Factors in Computing Systems (CHI '25) (Honorable mention award - top 5% out of more than 5,000 submissions)
Linguistic Alignment Predicts Learning in Small Group Tutoring Sessions
Dorothea French, Robert Moulder, Kelechi Ezema , Katharina von der Wense, Sidney K. DMello
Empirical Methods in Natural Language Processing (EMNLP) 2025
AfriVox: Probing Multilingual and Accent Robustness of Speech LLMs
Busayo Awobade, Mardhiyah Sanni, Tassallah Abdullahi, Chibuzor Okocha, Kelechi Ezema, Devendra Deepak Kayande, Lukman Enegi Ismaila, Tobi Olatunji, Gloria Ashiya Katuka
European Chapter of the Association for Computational Linguistics (EACL) 2026
AfriMMT-EA: Multi-domain Machine Translation for Low-Resource East African Languages
Naome A Etori, Kelechi Ezema, Nathaniel Romney Robinson, Davis David, Alfred Malengo Kondoro, Elisha Ondieki Makori, Michael Samwel Mollel, Maria Gini
European Chapter of the Association for Computational Linguistics (EACL) 2026
Selected Talks/Presentations
Robust debaising of Automatic Speech Recognition Systems
Presented research finding on accurate ASR systems for NLP applications at the 2025 Summit for AI Institutes Leadership (SAIL) inaugural Graduate Student and Postdoc Workshop, discussing implications for educational technology and inclusive design.
Evaluating the Pedagogical and Cultural Alignment of Rori AI
Presented internship findings to the Anthropic and Rising Academies teams regarding the evaluation of Rori AI. The study focused on the pedagogical utility and cultural appropriateness of the Claude-based system across four African languages.
Examining and Mitigating the Cascading Effects of Bias in Automatic Speech Recognition in Automatic Feedback Interfaces
Presented research findings on mitigating bias in ASR systems at Winter Text and Discourse (WT&D) conference, discussing implications for educational technology and inclusive design.
Innovating the Next Generation of STEM
Co-anrchord a graduate student workshop with Dr. Christina Alston focused on Innovating the Next Generation of STEM and practical approach to navigating gradsuate school
Selected Experience
Graduate Research/Teaching Assistant
- •Conduct research on the LEVI project, exploring the integration of human tutoring and AI to foster rich pedagogical discussions and collaborative learning
- •Assist in the Computer Science department, facilitating undergraduate lab sessions in Software Development and Introductory Programming classes
Researcher Internship
- •Analyze chat conversations and learner interactions with AI bots (Rori and Tari) to evaluate user behavior and gain insights into platform effectiveness.
- •Develop and refine prompt engineering strategies to enhance AI-generated math lessons and quizzes for better engagement and learning outcomes
Software Developer
- •Developed a backend dashboard for a series of educational platforms used in public universities across southeastern Nigeria
- •Implemented responsive frontend designs to improve user engagement and enhance the performance of educational applications and online resources
Professional Involvement
International Conference on Artificial Intelligence in Education (AIED 2026)
- ●Appointed as a Program Committee member for the main track of the 2026 International Conference on Artificial Intelligence in Education. Served as a peer reviewer for academic submissions and assisted with organizational logistics for the conference.
African Graduate Students' Association (AGSA) - CU Boulder
- ●Elected as the pioneering president of AGSA, leading the official registration of the organization as a Recognized Student Organization (RSO) at CU Boulder, while spearheading initiatives to foster community engagement and provide a sense of home for members
Society for Text & Discourse
- ●Volunteered as a local conference organizer for the annual Winter Text and Discourse Conference (WT&D), actively participating in logistics to ensure a successful event, receiving high commendations from participants
Office of Graduate Access & Retention
- ●Mentored rising juniors and seniors participating in the 10-week SMART program, offering guidance and innovative insights to emerging researchers
Federal Road Safety Corps (FRSC) Community Development Service (CDS)
- ●Elected president of the NYSC community group, leading road safety awareness programs and facilitating outreach to local communities on safety practices and the importance of integrating safety strategies into daily life
Miscellaneous Projects
Quantizations of Pre-trained models
A final class project focused on the quantization (parameter reduction) of five pre-trained models (BERT, DeBERTa, ELECTRA, etc.), resulting in a condensed model capable of running efficiently on mobile devices.
Gender and Racial disparities in Llama 3 and 3.2 Models
A class project analyzing gender and racial bias in Meta’s Llama 3 and 3.2 models. Evaluation was conducted at both the encoding stage (using WEAT method) and output stage (through prompt engineering).
Fine-Tuning of Phi-4 ASR Model
A research project focused on fine-tuning the SOTA Microsoft Phi-4 Automatic Speech Recognition (ASR) model on accented Africa speech, which contributes to improving the accuracy of speech-to-text applications.
Evaluating the Medical Safety of ASR Models
Developing a safety taxonomy and evaluation framework for ASR models in medical contexts; analyzing the effects of transcription errors across diverse accents and cross-architectural model families to identify potential clinical risks.
AfriVox: Multilingual Speech Recognition
Research project investigating accent robustness in speech language models, focusing on African languages and dialects for inclusive AI systems.
Online Result Processing System
A comprehensive web-based system designed for managing and processing academic results in secondary schools, featuring automated grade calculations and report generation.
Real Estate Website
Full-stack real estate solution with payment integration, inventory management, and user authentication built with React and Node.js.