How this site is put together

One page. A sticky identity spine on the left, content bands on the right, a putting green at the foot. No theme — the files below are the whole site.

index.md                   opening paragraphs + the service line
_data/research.yml         research interests
_data/teaching.yml         teaching appointments
_publications/*.md         one file per paper (selected = numbered ref,
                           featured = full entry too)
_layouts/spine.html        the page
_includes/site/spine.html  the left column
_includes/site/green.html  the putting green
_includes/site/icon.html   inline SVG icons, looked up by name
assets/css/site.css        every style on the site
assets/js/green.js         the putt
images/papers/*            one thumbnail per paper

Adding a paper

Drop a file in _publications/, named YYYY-MM-DD-slug.md. The date only controls the order — newest first — so use the preprint date.

---
title: "Paper Title"
date: 2026-02-14
key: slug                               # the citation handle, see below
link: https://arxiv.org/abs/...         # where a citation number points
venue: "NeurIPS 2026 · Spotlight"      # shown small and plum above the title
authors: "<b>Steffi Chern</b>, A Coauthor, B Coauthor"   # <b> your own name
image: /images/papers/slug.svg          # optional, see below
selected: true                          # include it in the numbered refs
featured: true                          # and show it under Selected publications
links:
  - { label: Paper, url: https://arxiv.org/abs/..., icon: paper }
  - { label: Code,  url: https://github.com/...,    icon: code  }
  - { label: Site,  url: https://...,               icon: site  }
---

icon: accepts paper, code, site. A paper with no image gets a plain tinted tile, so the column never breaks.

The thumbnails

images/papers/ currently holds placeholder marks I drew — they are not figures from the papers. Replace each one with a 4:3 crop of that paper’s Figure 1, about 420×315, and point image: at it. That reads far better than an abstract mark.

Changing the rest

  • Name, role, location, links, photo → the author: block in _config.yml
  • Research → _data/research.yml, one body: per paragraph (currently a single sentence that carries its own 🧠 lead-in). Cite a paper inside a sentence by writing [[id]], where id is the id: of a file in _publications/ — it renders as a linked [3] right there, so each clause can point at the work it describes. The numbers are handed out in the order the citations appear in the text, so rewriting or reordering the sentence renumbers everything and they always climb as you read. A paper with selected: false cannot be cited
  • Teaching → _data/teaching.yml (newest first)
  • Reviewing line, the publications lead-in → front matter in index.md

The old template

The academicpages files are still on disk but every one of them is listed under exclude: in _config.yml, so none of it is built or published — no placeholder papers, no sample blog posts, no /cv/ or /talks/ pages. Delete those directories whenever you want; just drop the matching lines from exclude: when you do.

Running it locally

bundle install
bundle exec jekyll serve

Two names you cannot use in a paper’s front matter

key: and link: are spelled that way on purpose. Jekyll defines id and url on every collection document, and those definitions win over front matter, so a paper written with id: and url: hands Liquid the document’s own path instead — the citations silently stop resolving and [[slug]] shows up as literal text on the page.

Research

  • 🧠 My current research interests include 1) making AI systems trustworthy and robust [1], [2], [3], [4], and 2) building multimodal AI models that can perceive and reason across images [5],[6] and videos [7],[8].

Selected publications

Preprint · 2025

Thinking with Generated Images

Ethan Chern, Zhulin Hu, Steffi Chern, Siqi Kou, Jiadi Su, Yan Ma, Zhijie Deng, Pengfei Liu

AAAI AI4Research 2025 · Oral

Can Large Language Models be Trusted for Evaluation? Scalable Meta-Evaluation via Agent Debate

Steffi Chern, Ethan Chern, Graham Neubig, Pengfei Liu

COLM 2025

FacTool: Factuality Detection in Generative AI — A Tool Augmented Framework for Multi-Task and Multi-Domain Scenarios

I-Chun Chern, Steffi Chern, Shiqi Chen, Weizhe Yuan, Kehua Feng, Chunting Zhou, Junxian He, Graham Neubig, Pengfei Liu

Teaching

  • Fall 2026 Teaching Assistant, Mathematics for Machine Learning
    University of Pennsylvania
  • Spring 2024 Teaching Assistant, Advanced Methods for Data Analysis
    Carnegie Mellon University
  • Fall 2023 Teaching Assistant, Probability Theory
    Carnegie Mellon University