RESEARCH MENTORING · IRVINE, CA · ONLINE

Real AI research, mentored one-on-one.

A 16-week program where high school students design, run, and write up their own NLP/LLM research project, guided start to finish by an EMNLP-published AI scientist and university AI lecturer.

EMNLP 2026 · first author1,700+ citationsAI Lecturer · Kookmin University program at Westcliff UniversityEx-Ericsson Senior Research ScientistPh.D. · U. of Manitoba

01 // YOUR MENTOR

Mentored by someone who publishes, not a matched tutor

Publishing in NLP now

First author of an EMNLP 2026 main-conference paper on retrieval-augmented small language models.

Has taken student research to a top venue

That EMNLP paper grew out of the student research group I lead on on-device language models.

Teaches AI for a living

Teaches LLMs and applied AI to university students in Kookmin University's global program at Westcliff University in Irvine. NVIDIA Certified Deep Learning Instructor.

Industry-grade research

Led a 10-person generative-AI research team at Ericsson. 1,700+ Google Scholar citations.

02 // WHO IT'S FOR

Built for students who want to do the work

  • →High school students (grades 9–12) who want real research, not another course.
  • →Students aiming for science fairs, AI competitions, or an authentic research story for college applications.
  • →Comfortable with basic Python, or willing to finish a two-session bridge first.
  • →Ready to put 4–6 hours a week into their own project.

Advanced middle school students are considered after an interview.

03 // THE 16-WEEK PROGRAM

From a question to a finished paper

Four phases, one mentor the whole way, and a progress report to parents every four weeks.

Length
16 weeks
1:1 session
60 min weekly on Zoom, recorded
Async review
Code & writing feedback within 48 hrs
Parent reports
Weeks 4, 8, 12 and 16
Student time
4–6 hrs / week
Language
English or Korean
  1. 01 · WEEKS 1–4

    Foundations & the research question

    The student can say what they are studying and why.

    • ·Setup: Python, Google Colab, GitHub
    • ·How LLMs work: tokens, embeddings, transformers, prompting
    • ·How to read papers; literature review of 5–8 papers
    • ·A testable question and hypothesis

    → One-page research proposal

  2. 02 · WEEKS 5–8

    Data & baseline

    A first, honest number to beat.

    • ·Find or build a dataset; licensing and privacy
    • ·Reproduce an existing method as the baseline
    • ·Design the evaluation: metrics, test sets, LLM-as-judge pitfalls
    • ·Mid-point review and course correction

    → Baseline results + mid-point slides

  3. 03 · WEEKS 9–12

    Experiments & analysis

    The student's own idea, tested properly.

    • ·Implement the student's proposed method
    • ·Ablations: which part actually helps?
    • ·Error analysis on real failure cases
    • ·Robustness check on a second dataset or model

    → Final results tables

  4. 04 · WEEKS 13–16

    Paper & presentation

    Work that is ready to submit and present.

    • ·Figures, tables and outline
    • ·Full draft, written by the student with mentor feedback
    • ·Revision and a reproducible public GitHub repo
    • ·Final talk rehearsal and a submission plan

    → 4–8 page paper, repo, poster or slides

04 // EXAMPLE TOPICS

NLP & LLM research a high schooler can own

Every topic runs on free Colab and small open models within 16 weeks. Students can also bring their own idea.

  • Measuring hallucination in small LLMs

    Intro

    Which kinds of questions make 1B–8B models invent answers, and how does that change with model size?

  • Prompt sensitivity in math word problems

    Intro

    Does changing a single word flip the model's answer? Which edits break it most?

  • Korean–English code-switching

    Intermediate

    How well do LLMs understand and translate mixed Korean–English (Konglish) text? A topic where bilingual students have an edge.

  • How reliable are AI-text detectors?

    Intermediate

    Do detectors flag writing by non-native English speakers as AI-generated more often?

  • Retrieval-augmented QA on local knowledge

    Intermediate

    Over a narrow corpus like a school course catalog, why does RAG find the answer but fail to state it exactly? Connected to my EMNLP work.

  • Fairness of LLM graders

    Intermediate

    When an LLM grades essays, does the score shift with the writer's name or dialect?

  • On-device LLMs: the quantization trade-off

    Advanced

    How much accuracy does 4-bit or 8-bit quantization cost, and how much faster does a phone get?

  • Tool use in small-model agents

    Advanced

    When do small models call a calculator or search tool correctly, and how do they fail?

05 // WHAT YOU LEAVE WITH

Work that stands on its own

  • ✓A 4–8 page research paper in conference format
  • ✓A public GitHub repo with code, a data card and a reproduction guide
  • ✓A poster or presentation deck
  • ✓A submission roadmap matched to the student's level and timeline
  • ✓A mentor evaluation describing the work the student actually did, on request

Where students can submit

  • ·High school research journals (e.g., Journal of Emerging Investigators)
  • ·arXiv preprints
  • ·Student research workshops and high school tracks at AI conferences, when offered
  • ·OC Science & Engineering Fair → ISEF, Regeneron STS
  • ·USA AI Olympiad (USAAIO), Congressional App Challenge

No publication, award or admission outcome is guaranteed. What is guaranteed is the quality of the research process.

06 // RESEARCH INTEGRITY

The student does the research

Code, experiments and writing are done by the student. I guide, review and give feedback.
Authorship reflects real contribution. I do not write papers for students.
I do not write college essays. I help students describe their research accurately.
Any AI-tool use is disclosed and logged in the research notebook.

07 // FAQ

Questions parents ask

Does my child need coding experience?+

Basic Python (variables, loops, functions) is enough. If they are not there yet, we start with two bridge sessions before Week 1.

Will my child get published?+

That depends on the student's work and on reviewers, so I never promise it. The program is built so the result is strong enough to submit, and we pick venues that fit.

How is this different from large mentoring programs?+

You work directly with the researcher who publishes in this field, not a matched graduate student, and the same mentor stays with you for all 16 weeks.

Are sessions in English or Korean?+

Either. Sessions can be in English or Korean, and parent reports come in the language you prefer.

We're not in Irvine. Can we still join?+

Yes. Everything is online. Sessions run on Pacific Time, and weekend slots also work for families in Korea.

What does it cost?+

Tuition is shared during the free consultation, after we've talked about your child's goals and level.

08 // START HERE

Start with a free 30-minute consultation

Tell me your child's grade, interests and goals. We'll talk about whether research is the right fit and what a realistic project looks like.

Email to book a consultation

donghok4@gmail.com

© 2026 DR. HENRY KANG · HOME · GOOGLE SCHOLAR