Human–Computer Interaction • Machine Learning • Technology Policy

I design human-centered ML systems — decision support people can understand, challenge, and trust.

I’m Yogesh Luitel — a computer science instructor and early-career researcher. Technology is never neutral. We research what it does to people.

Projects Research focus Open CV
Current
CS Instructor (Grades 11–12)
Xavier International College • Kathmandu
Research
Human–AI interaction
Trust calibration • uncertainty • evaluation
Policy lens
Accountable deployment
Privacy • auditability • governance

About

I work at the boundary between research and practice: human–AI interaction, empirical ML evaluation, and technology policy for accountable deployment.

Short bio

I’m interested in how people make decisions with ML systems: what they see, what they trust, and what they can contest. I focus on human-centered design patterns (uncertainty cues, explanations, feedback loops) paired with evaluation methods that go beyond accuracy.

My research hook

Turning predictions into decision support people can calibrate: evidence + uncertainty + accountability — not just scores.

Research interests

Human–AI interaction Trust calibration Uncertainty-aware UX Explainability Empirical evaluation Robustness & failure modes Transparency & governance Privacy-by-design
  • HCI: interface patterns for sensemaking, feedback, and contestability.
  • ML: evaluation, subgroup errors, robustness, interpretability.
  • Policy: accountability constraints for real institutional deployments.

Selected projects

Polished, research-aligned projects: interface design + evaluation + real-world constraints.

HCI • XAI

Human-Centered ML for Programming Misconceptions

Decision Support

Teacher-facing decision support that surfaces likely misconceptions with uncertainty cues and explanations — designed to reduce over-trust and preserve instructional agency.

  • Interface: uncertainty signaling + evidence views + easy override
  • Evaluation: behavior metrics (time-to-decision, correction patterns)
  • Goal: support sensemaking, not automated grading
Python Human-in-the-loop Explainability Evaluation
GitHub
Robustness • Reproducibility

ML Evaluation & Interpretability Toolkit

Audit-ready

A reproducible evaluation pipeline that turns “good accuracy” into deployable evidence: subgroup errors, robustness checks, and interpretability reporting.

  • Evaluation: metrics + subgroup slices + failure modes
  • Interpretability: model-agnostic explanation views
  • Use case: institutional review & governance
scikit-learn SHAP Error analysis Reporting
GitHub
Empirical ML • NLP

GenAI Usage & Learning Outcomes Study

Methods-first

Empirical analysis of how generative AI usage patterns relate to learning outcomes using structured data + text features, emphasizing careful claims and reproducible methodology.

  • Data: usage logs + text signals
  • Methods: interpretable modeling + effect-size-grounded validation
  • Outcome: manuscript in preparation (share on request)
TF–IDF Feature engineering Stat tests Reproducibility
Request details
If a project is private, I can share a short demo clip or 1-page writeup by email.

Research

Where I want to go next: HCI + ML evaluation + policy-aware deployment.

Current focus

  • Human–AI interaction: decision-support UX, trust calibration, contestability.
  • Human-centered ML: uncertainty cues, explanation interfaces, feedback loops.
  • Empirical evaluation: subgroup errors, robustness, failure modes, reproducibility.
  • Technology policy: transparency, privacy, governance constraints.

Drafts & directions

  • GenAI usage → outcomes: empirical ML + NLP methodology (in preparation).
  • Accountability for ML-assisted decisions: uncertainty + oversight design patterns.
  • Low-resource policy constraints: privacy-by-design and evaluation standards.
Reach out about collaboration

Experience

Teaching + research + building — grounded in real users and constraints.

2024–Present
Computer Science Instructor • Xavier International College

Teach Python, DBMS, and software development; lead labs and supervise projects; design digital learning materials and workflows.

Python DBMS Mentoring Learning workflows
2023–2024
Undergraduate Research Assistant • Phoenix College of Management

Studied adaptive learning systems and built ML-based student performance analysis to identify conceptual weaknesses; supported research design and evaluation.

Learning analytics Mixed methods ML evaluation Adaptive learning

Contact

If you want to talk research, collaboration, or opportunities — email is best.

Links

GitHub LinkedIn

I’m especially interested in opportunities where interface design, evaluation rigor, and governance matter as much as the model.

CASE STUDY

Human-Centered ML for Programming Misconceptions

Teachers stay in control — ML assists, explains, and invites correction

Decision support that encourages verification rather than blind acceptance.

Paradigm
Human-in-the-loop
UX
Uncertainty + XAI
Evaluation
Behavior metrics

Design

HCI
  • Confidence/uncertainty cues without false precision
  • Evidence views that support sensemaking
  • Easy override and feedback capture

Evaluation

Empirical
  • ML On vs ML Off (pilot) with interaction logging
  • Time-to-decision + correction patterns
  • Checks for over-reliance behavior
CASE STUDY

ML Evaluation & Interpretability Toolkit

From “good accuracy” to deployable evidence

Reproducible evaluation that supports institutional oversight and governance.

Checks
Subgroups + robustness
Output
Auditable reports
XAI
Model-agnostic

Pipeline

Reproducible
  • Standardized dataset prep → training → evaluation → reporting
  • Failure modes and subgroup slices for fairness-adjacent risk
  • Robustness checks to surface brittleness early

Governance angle

Policy
  • Supports “should we deploy?” decisions, not only “is it accurate?”
  • Pairs with transparency requirements and documentation
CASE STUDY

GenAI Usage & Learning Outcomes Study

Empirical ML that avoids hype

Interpretable modeling, careful claims, reproducible methodology.

Data
Structured + text
Methods
Interpretable
Status
In preparation

Method

Empirical
  • NLP features (TF–IDF + statistical signals) for interpretability
  • Effect-size-grounded validation; reproducible experiments
  • Focus on failure modes and alternative explanations

Implications

Policy
  • What institutions should measure before adopting GenAI at scale
  • Transparency and responsible-use constraints