VOL. 01 · HONG KONG · 2026

FRONT PAGE // SEPTEMBER 2026HONG KONG

Photo of Eman Ansar

researcher, note-taker, occasional poem-hoarder

[research notes on vision, behavior, language, and cognition]

Eman Ansar

Researching human reasoning, decision-making, and AI that helps people think independently.

I’m an M.Phil. (Research) student in Computer Science & Engineering at HKUST, where I study what visible behavior can reveal about thinking in expert chess. My work sits at the intersection of computer vision, HCI, and visualization, with earlier detours through Arabic NLP and resource-efficient speech models. Before this, I studied Computer Science at Carnegie Mellon, where I also taught and did research in vision and language.

THE INVESTIGATIONS // CURRENT FILES

Featured Investigations

Current questions about visible behavior, structured choice, and the meanings carried by visual form.

HOT FILE // 01UNDER RESUBMISSION

No Single Tell

What can visible behavior actually tell us about thinking?

I align long-form chess video with human behavioral annotations, game information, clocks, and engine-derived decision context to study visible behavior over time.

open the field file ↗ open PDF in Drive ↗
Two chess players concentrating across a tournament board
World Championship chess // concentration, posture, clocks, silence.

UNDER INVESTIGATION // 02

COGSCI 2027 · MANUSCRIPT IN PREPARATION

Choice Complexity

What makes a decision difficult when several options are all annoyingly plausible?

We model the full choice set and ask how its structure relates to decision time and decision quality.

read the field file ↗
A hand hovering over a chessboard beside a clock
the problem with having more than one reasonable move

WORLD DESK // 03

ACM CHI · LATE-BREAKING WORK IN PREPARATION · IEEE VIS 2027 FULL PAPER

Multilingual Visualization

When a chart crosses a language boundary, what exactly has to travel with it?

I study how people interpret, explain, and recreate visualizations across linguistic and cultural contexts.

field file ↗ PDF ↗
Printed Arabic, Chinese, and European charts arranged as multilingual visualization research materials on a desk
apparently the bars are universal. everything around them has opinions.

EARLIER FILES // FORMATIVE DETOURS

From the Archive

Earlier projects that taught me how to build datasets, distrust preprocessing defaults, and eventually ask more human questions.

ORIGIN STORY // 05

IEEE VCIP · 2023–24

Thermal Vision for Solar Panels

Published comparison of raw thermal imagery, normalization, and shadow exclusion for solar-panel feature detection

Sophomore me, several notebooks, and a very sunny CV problem.

Benchmarking classical computer-vision methods on close-range thermal imagery.

open field file ↗ paper PDF ↗

THESIS DESK // 06

CMU HONORS THESIS · 2024–25

Speech & Audio LLM Pretraining

Research poster for the Speech and Audio LLM pretraining honors thesis

The lab where everyone seemed to speak in loss functions.

Resource-efficient representations, synthetic supervision, and my senior honors thesis.

open field file ↗ thesis PDF ↗ poster PDF ↗

COURSE PROJECTS // SYSTEMS // THINGS THAT ESCAPED THE NOTEBOOK

Selected Technical Detours

Go Small or Go Home project poster

GENERATIVE AI · CARNEGIE MELLON · NOV—DEC 2024

Go Small or Go Home

Lightweight text and image generation for mobile devices using quantization, pruning, and knowledge distillation.

view full-resolution poster ↗
Research poster for deep learning and microblogs for emergency resource allocation

QSIURP · CARNEGIE MELLON · MAY 2024—JAN 2025

Deep Learning for Emergency Resource Allocation

A multi-task pipeline for classifying disaster-related microblogs and extracting resource, quantity, and location information from the 25-class TREC-IS dataset.

project PDF ↗ poster PDF ↗
Diagram page from the Spectral Adaptive FLORA course report

COMP6211J · PROPOSED METHOD + EXPERIMENTAL PLAN

Spectral Adaptive FLORA

A literature-grounded extension of FLORA using adaptive projection resampling, multi-scale gradient decomposition, and spectral consistency.

read report ↗
Sensor glove and electronics for the ASL Automated Interpreter

EMBEDDED SYSTEMS · ROBOTIC PROTOTYPE

ASL Automated Interpreter

A sensor-equipped glove that maps hand configurations to ASL letters, displays output on an LCD, and supports sentence composition.

GitHub + prototype ↗
Matrix multiplication benchmark for the Apple-Silicon Needle backend

GRADUATE ML SYSTEMS · UNDERGRADUATE PROJECT

Apple-Silicon Acceleration for Needle

A Metal backend implementing core tensor operations and benchmarking Apple M-series acceleration against the CPU backend.

presentation ↗ benchmark ↗

ACADEMIC BACKGROUND

Education

A Levels · 4 A*s

2021 · Mathematics · Chemistry · Biology · Physics

High Achiever distinction.

WHERE THE WORK HAPPENED

Research Rooms

2022
✦ Most Promising Research Proposal

this one may have accidentally encouraged the entire situation.

2023

May 2023—Jul 2025

IMPAQT Lab · Carnegie Mellon

Undergraduate Researcher · Eduardo Feo-Flushing

Thermal computer vision, solar-panel imagery, and the first long route from messy data to a published paper.

open the origin story ↗
2024

May 2024—Jan 2025

Summer Fellowship Program · Carnegie Mellon

Undergraduate Research Fellow · Gianni Di Caro

Deep learning for crisis-information extraction and resource planning.

May 2024—May 2025

ML & Signal Processing Lab · Carnegie Mellon

Honors Thesis Researcher · Bhiksha Raj

Speech and audio LLM pretraining through compressed representations and synthetic supervision.

open the thesis field file ↗
2025

Apr—Aug 2025

National Research Fund · Arabic Writing Assessment

Research Assistant · Houda Bouamor (Qatar)

PAIR-Net, ALiF-Net, prompt-agnostic Arabic assessment, and cross-lingual augmentation.

open the field file ↗
Doha / Pittsburgh → Hong Kong ✦ Fully Funded Postgraduate Research Studentship · HKUST · Aug 2025

packed several years of research lore into approximately too many suitcases.

Oct 2025—Present

DataVisards Lab · HKUST

Postgraduate Research Assistant · Arpit Narechania

Current work across embodied cognition, human-centered AI, and multilingual visualization.

01No Single Tell 02Choice Complexity 03Multilingual Visualization

THE CLASSROOM THREAD

Teaching Rooms

CARNEGIE MELLON2024—2025

11-785: Introduction to Deep Learning

Teaching Assistant · three semesters · Pittsburgh → Doha

I spent three semesters as an undergraduate TA for CMU’s flagship graduate deep-learning course, mentoring graduate students, co-developing assignments with Bhiksha Raj, and supporting work across CNNs, sequence models, Transformers, and LLMs.

HKUSTSPRING 2026

COMP 4211: Machine Learning

Teaching Assistant

I independently led lab tutorials and built instructional notebooks for core machine-learning methods, emphasizing implementation students could reason through rather than treat as magic.

CARNEGIE MELLONSPRING 2025

15-288: Machine Learning in a Nutshell

Head Teaching Assistant

I led weekly recitations, mentored capstone projects, and delivered guest lectures on Transformers, self-attention, and generative AI—moving from supporting instruction to helping shape how the material was taught.

CARNEGIE MELLONJAN—MAY 2024

15-110: Principles of Computing

Teaching Assistant · Giselle Reis

I led recitations and office hours on Python, algorithmic reasoning, and introductory computer science.

THE LONG-TERM FORECAST

Why I keep coming back to it

Research gives me the particular joy of being confused for a very long time and then, occasionally, less confused. Teaching gives me something different: watching the exact second something stops being impossible for somebody else.

I believe ability is much less fixed than opportunity makes it look.
drumroll, please 13,407,069 What do you think this number counts? reveal the answer ↓

girls aged 5–16 reported out of school in Pakistan

Census 2023, via State of Children in Pakistan 2024 ↗ not a measure of ability.

I keep trying to think of a clever comparison for 13,407,069, and I actually don’t want one.

It is already big enough.

Those are not thirteen million failed learners. They are thirteen million lives in which access, time, safety, money, expectations, teachers, or simply permission may have failed first.

I don’t think I want to choose between research and teaching.

Part of that is personal. I grew up around women—cousins, family, girls I knew closely—who were every bit as sharp as I was. Some were sharper. But intelligence is not the same thing as choice. I watched lives become narrower because the menu of acceptable futures was narrow: geography, money, family expectations, safety, timing, or simply whether somebody was allowed to keep going.

My own life contains an uncomfortable amount of geographical luck. I grew up in Qatar. I had parents who kept saying yes. I landed at Carnegie Mellon. I found labs with compute, funding to travel, professors who answered the extra question, and mentors willing to spend hours on a student who did not yet know enough to deserve those hours on paper.

I am, embarrassingly, a composite object made out of other people’s patience.

Eduardo let sophomore-me into a research lab when I mostly had curiosity and Jupyter notebooks. Bhiksha sat with me in Pittsburgh and walked through papers paragraph by paragraph until I learned how researchers actually read. Other mentors gave me second attempts, longer explanations, compute, introductions, and room to be unfinished.

There is research behind the feeling, too. One longitudinal study associated having a school-based natural mentor with a 9.4 percentage-point increase in college attendance, with the largest effects among students from lower socioeconomic backgrounds. A recent meta-analysis of faculty–student relationships similarly examined their relationship with achievement and persistence in higher education.

That is probably why Benjamin Bloom’s idea of mastery learning has lodged itself so firmly in my head. In Learning for Mastery, aptitude is treated partly as the amount of time a learner needs to reach mastery—not simply a fixed ceiling on what they can learn.

I think about that whenever we are too quick to call somebody “not a math person,” “not technical,” or “not research material.”

And the educational divide does not disappear when you reach a university. It just gets quieter. One student gets an H200, conference travel, people who read three drafts, and the freedom to run the experiment again. Another gets a laptop, a cloud-credit limit, and documentation.

Both can be brilliant. One simply gets more iterations.

I want to spend part of my life giving people more iterations. More time. Better explanations. A second attempt. Access to the machine. A seat in the room. Someone willing to say, show me exactly where it stopped making sense.

That is the kind of educator I think I want to become.

BOOKS IN THE MARGINS

The ideas underneath the instinct

long-term forecast: educator behavior remains highly likely.

LECTURES // TALKS // DEFENSES // OCCASIONAL MICROPHONE POSSESSION

At the Podium

APR 2026 · RESEARCH TALK

Three Minute Thesis

HKUST · First Runner-Up

Presented my M.Phil. research on visible behavior and decision-making in elite chess, compressing a fairly unreasonable amount of HCI, computer vision, and chess into three minutes.

watch the 3MT talk ▶

SPRING 2025 · GUEST LECTURES

Guest Lecturer & Head TA · 15-288

Carnegie Mellon

Delivered three lectures on Transformers, self-attention, and generative AI while leading weekly recitations and mentoring student capstone projects.

lecture slides PDF ↗course record ↗

SPRING 2025 · THESIS DEFENSE

Undergraduate Honors Thesis Defense

Carnegie Mellon

Defended my honors thesis on compute-efficient pretraining for speech and audio language models, spanning compressed representations, synthetic supervision, and speech-error modeling.

open the thesis field file ↗

MAY 2025 · RESEARCH SYMPOSIUM

Meeting of the Minds

Carnegie Mellon

Presented my Summer Fellowship Program research on deep learning for crisis-information extraction and resource planning.

symposium record ↗

ALSO BEHIND A MICROPHONE

Recurring instruction & applied presentations

  • 2023Summer Undergraduate Research Apprenticeship Symposium · thermal computer vision
  • 2024—2511-785: Introduction to Deep Learning · three semesters · Pittsburgh → Doha
  • 2026COMP 4211: Machine Learning · lab instruction · HKUST
  • IndustryMicrosoft Internship Innovation Competition · multilingual captioning prototype · 2nd Place

INDUSTRY // APPLIED WORK

Industry Rooms

Microsoft

INTERNSHIP

Artificial Intelligence Intern

Led a five-person team building a real-time multilingual captioning prototype using Azure and Power Apps, integrating live transcription, translation, and accessibility features into one end-to-end workflow.

Snoonu

SUMMER 2024

Backend Software Engineer Intern

Built shopper-operations backend services using .NET 8, MongoDB, and Docker, supporting modular order and inventory workflows through unit-tested APIs and service architecture.

Mety Technology

MAY—JUL 2024 · UNITED STATES

AI Product Development & Marketing Intern

Automated newsletter and content-production workflows and applied LDA and sentiment analysis to market research, audience segmentation, and product-marketing decisions.

Deloitte D-180

MAY—DEC 2024 · DOHA / REMOTE

D-180 Mentee

Participated in Deloitte’s D-180 mentorship program, gaining exposure to consulting, professional practice, and industry problem-solving alongside university and research work.

PAPERS // MANUSCRIPTS // THINGS CURRENTLY ANNOYING REVIEWERS

Printed, Pending, and Still Brewing

  1. 01

    No Single Tell: Studying Context-Dependent Visible Behavior During Extended Decision-Making in Elite Chess

    Under resubmission

    paper PDF ↗
  2. 02

    Multi-locale Chart Interpretation

    17-participant study · ACM CHI late-breaking work in preparation · IEEE VIS 2027 full paper planned

    paper PDF ↗
  3. 03
  4. 04
  5. 05
  6. 06

    VerLM: Explaining Face Difference Using Natural Language

    Research manuscript

    manuscript PDF ↗
  7. 07

    Harnessing Deep Learning and Microblogs for Real-Time Resource Allocation in Emergency Scenarios

    Research manuscript

    manuscript PDF ↗

LIFE // POEMS // HCI FOR NORMAL PEOPLE

Three Desks, Same Brain

I write in three registers: essays about living, poems about feeling, and research explainers about how people and technology make sense of one another.

researcher · explainer · poet · person with too many tabs

Life // Notebook

Notebook

Essays and field notes on life, faith, ambition, womanhood, becoming, and the little arguments I’m still turning over.

read the life notes ↗

New publication HCI // Research Explainers

Eman Explains

Interesting research, complicated ideas, and things I can’t stop thinking about—explained like a normal person. I promise.

We’ll learn A through Y, then get to Z together.

I read papers, fall into research rabbit holes, get confused, figure them out, and explain what I wish someone had told me at the beginning. Difficult ideas do not need to be dumbed down; sometimes they just need someone willing to build the staircase.

read Eman Explains ↗ about the publication ↗

OPEN LINE

Correspondence

Letters, questions, collaborations, and curious detours are welcome.

DIRECT LINE // HOVER TO REVEAL send a note ↗

with curiosity,