Academic profile

Mayesha Maliha Proma

Mayesha’s research focuses on human-AI interaction, AI-supported decision-making, user behavior, trust, educational technology, NLP, and the real-world evaluation of intelligent systems.

Human-AI InteractionAI-supported Decision-MakingUser Behavior & TrustEducational TechnologyNLP & Language ModelsReal-world Evaluation
Portrait of Mayesha Maliha Proma
Mayesha Maliha Proma

Japan / Bangladesh

Research interests

Human-AI interaction, trust, educational technology, NLP, and whether intelligent systems help people in the situations where they actually use them.

Read the statement

Current work

Full-time English teacher at Maple Inc. Independent research, product work, and MASc applications for Fall 2027. Earlier: research assistant under Professor Parag Kulkarni (November 2023 — March 2026).

Ongoing research

Education

BSc in Digital Business and Innovation with a concentration in Data Science and AI from Tokyo International University. Full-tuition scholarship, April 2022 — March 2026.

View education

Education

BSc in Digital Business and Innovation

Tokyo International University

Bachelor of Science with a concentration in Data Science and AI. Full-tuition scholarship.

  • Recipient of TIU’s 100% tuition-reduction scholarship for the undergraduate degree (2022–2026).
  • Also received the Monbukagakusho Honors Scholarship for privately-financed international students (JASSO), April 2022.
  • Studied digital business, data science, artificial intelligence, machine learning, and product development.
  • Research assistant under Professor Parag Kulkarni (November 2023 — March 2026).
  • Teaching assistant / student assistant for Foundations of Python and IT Project Management (August 2024 — July 2025).
  • Founded the TIU Data Science and Analytics community and the Bangladesh Student Association.
Full-tuition scholarshipData ScienceAIDigital Business

Professional Certificate in Data Science

HarvardX / edX

Completed the nine-course Data Science Professional Certificate (credential issued November 2025).

  • Statistics, R, and machine learning coursework later used in research design and product decisions.
HarvardXData ScienceStatistics

Secondary and higher secondary education

Viqarunnisa Noon School and College

Science background before undergraduate study in Japan. This period also included early student leadership and education-community work.

ScienceBangladesh

Current status

After completing her bachelor’s degree in March 2026, Mayesha has been teaching English in Saitama while continuing independent research, academic writing, and product work through The Abroad Company. She is preparing MASc applications for Fall 2027.

Research Interests

Mayesha’s research experience began with applied NLP and machine-learning research under Professor Parag Kulkarni at Tokyo International University. Her work included text summarization, knowledge graphs, human-resource text analysis, patent summarization, low-resource NLP, and machine-learning applications. Her current research direction extends from technical AI development toward human-AI interaction and real-world evaluation. She is interested in how people interpret, trust, and respond to AI-supported decision systems, and whether those systems improve task completion, user behavior, and meaningful outcomes.

Human-AI interaction and decision-making

How people interpret, trust, and act on AI-generated guidance when the decision is consequential.

Human-AI InteractionTrustDecision Support

Real-world evaluation of intelligent systems

Whether AI systems improve decisions, task completion, trust, and access to opportunity — not only benchmark scores.

EvaluationUser BehaviorEducational Technology

Language models and educational NLP

Writing assessment, retrieval-augmented generation, knowledge-graph applications, and personalized language-based tools.

NLPLLMsAESKnowledge Graphs

Research methods

Methods Mayesha has used in papers and product research. Product observation is not listed as a formal human-subjects study.

  • Controlled feature isolation
  • Shuffled-label controls
  • Random-boundary controls
  • Random-feature controls
  • Bootstrap confidence intervals
  • Ablation
  • Human evaluation
  • Surveys
  • Interviews
  • Usability testing
  • User studies
  • Post-use feedback
  • Randomized feature experiments

Research experience

Research assistant

Professor Parag Kulkarni’s Research Lab

Undergraduate research assistant in natural language processing, text mining, machine learning, knowledge graphs, and applied artificial intelligence.

  • Conducted undergraduate research under Professor Parag Kulkarni in natural language processing, text mining, machine learning, knowledge graphs, and applied artificial intelligence.
  • Participated in weekly research meetings, research planning, idea development, study design, and academic writing.
  • Contributed to research on extractive and abstractive summarization, sentiment analysis, trend analysis, knowledge graphs, and NLP for low-resource languages.
  • Co-authored published research on graph-based text summarization for human resource management tasks.
  • Co-authored research on augmenting patent summarization with large language models and knowledge graphs, evaluated using automatic metrics and human assessment.
  • Developed independent research experience by initiating and leading an automated essay-scoring project during the final year of the degree.
NLPMachine learningKnowledge graphsUndergraduate research

Independent researcher

Human-AI interaction and real-world evaluation

  • Continues independent research on how people use AI-enabled tools in real products.
  • Asks whether recommendations, explanations, personalization, and human support change decisions, trust, and task completion.
  • Does not treat product observation as a completed formal human-subjects study unless ethics procedures are later documented.
Human-AI interactionEvaluationNLP

Ongoing Research

Ongoing Research

Evaluating AI support inside study-abroad products

A research direction on whether recommendations, explanations, personalization, and human support change how people decide, trust a tool, finish a task, or ask for help. Her current thesis work designs a context-aware personalization framework for AI agents: short- and long-term memory, a planner that suggests the next step, and memory viewers users can see and edit. This is work in design and observation, not a completed formal human-subjects study.

Role

Independent researcher

Current stage

Active reading, question-forming, and experiment design

Methods

User observation · Product analytics · Feature experiments · Qualitative feedback

Human-AI InteractionUser BehaviorEducational Technology

Selected publications and manuscripts

View publication record
In Progress

Automated Essay Scoring with Availability Signals and Genetic Algorithms: The RubriQ Framework

Combines DeBERTa-v3-base, availability-inspired essay signals, and genetic-algorithm fusion. The genetic algorithm improved mean held-out QWK from 0.7545 to 0.7635 and beat the baseline on all six prompts.

Role

Lead researcher and first author

Methods

DeBERTa-v3 · Genetic algorithms · Availability features · QWK

Automated Essay ScoringNLPRubriQ
Manuscript in Revision

What Makes Essay Structure Useful for Automated Scoring? A Document-Engineering Study of Boundary Geometry and Rhetorical Transitions

A predictive isolation study of whether discourse-derived document structure adds value beyond essay text, surface features, and component counts on PERSUADE 2.0. The complete structural model improved held-out QWK from 0.8361 to 0.8488. The study used controlled feature isolation, shuffled-label controls, random-boundary controls, random-feature controls, and bootstrap confidence intervals. This is not a causal study.

Role

Lead researcher and first author

Methods

PERSUADE 2.0 · Feature isolation · Shuffled-label controls · Random-boundary controls · Bootstrap CIs · QWK

Automated Essay ScoringNLPDocument Structure
Published

Survey of Graph-Based Text Summarization for HRM Tasks

Reviews graph-based text summarization methods for human resource management tasks and synthesizes the main approaches, applications, and research directions.

NLPIR 2024 · Proceedings of the 8th International Conference on Natural Language Processing and Information Retrieval · pp. 380–387

Role

Co-author

Methods

Literature Review · Graph-Based Summarization · NLP · HRM

Text SummarizationGraph MethodsNLPHRM
Published

Augmenting Patent Summarization Using a Large Language Model with a Knowledge Graph

Combines LLaMA 3 with patent knowledge graphs and evaluates the generated summaries through ROUGE and human assessment. Human evaluators preferred the knowledge-graph summaries in most tested patent cases, although the baseline achieved higher ROUGE scores.

LKM@IJCAI 2024 · First International OpenKG Workshop: Large Knowledge-Enhanced Models · pp. 1–13

Role

Co-author

Methods

LLaMA 3 · Knowledge Graphs · Neo4j · ROUGE · Human Evaluation

Patent SummarizationLLMsKnowledge GraphsNLP

Writing & Research Notes

Mayesha’s public writing includes beginner-friendly essays on artificial intelligence and an evolving research notebook where she records papers, questions, experiment ideas, and reflections.

Published Essays

LEARNING IN PUBLIC

Short Blogs on AI & Technology

These essays document how Mayesha learned technical ideas by explaining them in simple language and connecting them to tools, experiments, and products she was building. Together, they trace a path from foundational data science concepts to neural networks, language models, and applied AI systems.

Research Notebook

AI & Education Reflections

An evolving research notebook containing paper notes, open questions, experimental ideas, and reflections from ongoing work with AI-enabled systems.

Open research notebook →

Teaching experience

Teaching assistant / Student assistant

Tokyo International University

Supported undergraduate information-technology courses.

  • Supported undergraduate courses in Foundations of Python, IT Project Management, and related information-technology coursework.
  • Assisted classes of up to 150 students from more than 120 countries.
  • Helped students with programming exercises, presentations, attendance, class activities, and course-related questions.
  • Adapted explanations to different levels of technical knowledge, communication styles, and learning needs.
  • Supported students in understanding Python fundamentals, project-management concepts, and practical technology applications.
TeachingAcademic supportPython

Product and professional experience

Full-time English teacher

Maple Inc.

  • Teaches English full-time to Japanese professionals and Japanese children.
  • Plans, delivers, and revises monthly lessons based on learner progress and classroom observation.
  • Uses games, activities, and adaptive instruction to support different ages, attention spans, proficiency levels, and learning styles.
  • Adjusts teaching methods based on student engagement and observed learning needs.
English teachingJapan

Co-founder and product lead

The Abroad Company

Co-founded The Abroad Company and leads products that support students through study-abroad and international applications. The co-founder role is dated October 2025. AbroadMates itself began earlier, in 2022.

  • Leads product strategy, user experience, research direction, and early-stage operations.
  • Launched AbroadMates, a study-abroad mentorship platform for paid 1:1 conversations.
  • Leads ApplicationMate, an AI-supported application-planning platform for deadlines, essays, and scholarships.
  • Leads an 11-person distributed team across Japan, the United States, Canada, Bangladesh, and Italy.
Co-founderProduct leadershipStudy abroad

Data analyst and content marketing intern

Guidable Inc.

  • Designed internal surveys, collected data, and used R for analysis.
  • Researched sources, drafted and edited articles, and published to WordPress with SEO settings.
InternshipData analysisContent

Human-AI product research

Mayesha studied how users navigated the products, where they became confused, when they stopped responding, when they requested human assistance, and how feature changes affected completion, bookings, subscriptions, tool usage, and feedback. She does not claim that the products improved school admissions; that outcome has not been measured here.

  • Approximately 50 AbroadMates users and mentors
  • 10 closed-cohort ApplicationMate applicants
  • 856 chatbot participants
  • Surveys
  • Interviews
  • Usability testing
  • User studies
  • Post-use feedback
  • Randomized feature experiments

Users

Approximately 1,500

Booked sessions

More than 2,000

Mentors

210

Universities

Approximately 2,700

Programs

Approximately 157,000

These are current approximate figures. Booked sessions are sessions, not users.

Remote collaboration

Mayesha leads an 11-person distributed team across Japan, the United States, Canada, Bangladesh, and Italy. The team uses a shared portal, holds two meetings each week, and works primarily through asynchronous coordination.

Mayesha reviews weekly task completion, work activity, hours, and efficiency to improve the team’s systems and workflow. This is operational leadership, not a formal human-subject study.

Products & Research Platforms

Mayesha co-founded The Abroad Company and works on products that support students through study-abroad decisions. The co-founder role is dated October 2025; AbroadMates began in 2022.

Parent Company

The Abroad Company

The Abroad Company is the parent company for international education and mobility products, including AbroadMates and ApplicationMate.

Role

Co-founder, October 2025 — present. Mayesha leads product direction, research, and connected services across the student journey.

Focus areas

  • Product strategy
  • Research
  • International mobility
  • Distributed-team operations

These products create real-world environments for studying how human and AI support shape decisions and outcomes.

Leadership and community engagement

Founder and former president

TIU Data Science and Analytics Club

  • Founded a student community for data science, analytics, and AI; later the department’s student association.
  • Built a level-based curriculum with faculty; more than 70 students completed the pathway.
FounderData scienceStudent leadership

Founder

Bangladesh Student Association of TIU

  • Founded the association because no Bangladeshi student community existed on campus.
  • Helped incoming students and organized cultural festivals for four years.
FounderCampus community

Founder and former president

Women Opportunities

Founded a community that shared educational, professional, scholarship, and leadership opportunities with women. The work is currently paused. She may return to it later.

CommunityLeadershipWomen

Founder

Superwoman: A She Learning Platform

The e-learning initiative she started so women could learn practical business skills.

FounderLearning

Divisional coordinator

English Olympiad

Dhaka divisional coordinator, April 2017 — September 2022. Head of Sponsorship, November 2018 — January 2022.

CoordinationStudent engagement

Educational content developer

National Curriculum and Textbook Board, Bangladesh

Developed assignments for Bangladesh’s online education and evaluation during COVID-19.

CurriculumBangladesh

Project coordinator

Worldwide Organization for Charity

Volunteered, then helped plan programs including an unfinished coding initiative for children.

Community service

Awards and scholarships

Scholarships and awards.

TIU 100% Tuition Reduction Scholarship

Tokyo International University

Full-tuition undergraduate scholarship covering the Digital Business and Innovation degree.

Monbukagakusho Honors Scholarship for Privately-Financed International Students

Japan Student Services Organization (JASSO)

Honors scholarship for privately-financed international students.

Leadership Award

English Olympiad

Leadership award from the English Olympiad.

Technical skills

Research methods

How Mayesha designs and evaluates work.

Feature isolation · Control conditions · QWK evaluation · Human evaluation · Surveys and interviews · Usability testing

AI and NLP

Models and techniques she has used in papers and products.

Transformers · DeBERTa · RAG · Knowledge graphs · AES · Embeddings

Product and systems

Building and observing tools people actually use.

User research · Information architecture · EdTech workflows · Remote operations

Tools

Day-to-day stack.

Python · R · PyTorch · MongoDB · Next.js · React · LaTeX · Git

Languages

  • English (fluent)
  • Bengali (fluent)
  • Japanese (advanced)
  • Arabic (basic script)