Align 2D resume content with 3D site

CLAUDE.md

@@ -67,9 +67,8 @@ copyrightable); copying her stylesheets is not.
   `hugo.yaml` → `title`.
 - Skill progress bars only render `{{ if gt $level 0 }}` — so **omitting `level`** in
   the author YAML removes them. No override needed.
-- The collection block defaults to **`count: 5`**, which silently hid the trailing
-  "to be continued" card. It is now explicitly `count: 0` (unlimited). Never let a
-  default cap hide entries.
+- The collection block defaults to **`count: 5`**. It is explicitly `count: 0`
+  (unlimited), so the synchronized Research list is never silently capped.
 
 ## Local toolchain — three version walls (all worked around in scratch)
 

README.md

@@ -31,15 +31,15 @@ Motion is disabled automatically for visitors whose OS asks for reduced motion.
 
 No template assumes how many entries a list holds. Adding a project is one folder
 under `content/projects/`; adding a skill is one line in `data/authors/me.yaml`.
-Lists end in a **visible** placeholder rather than a silent cap — the research
-collection is explicitly set to `count: 0` so nothing is ever hidden behind a
-default limit.
+The research collection is explicitly set to `count: 0` so all four synchronized
+research entries remain visible without a default cap.
 
 The **Publications** section renders even while it is empty ("In preparation"). It
 does not get deleted just because there is nothing in it yet.
 
-Upstream truth for all of this lives in `PROFILE.md` (in the private `pw-meta`
-repo), projected into `profile-readme/data/profile.json`.
+Identity and skill data originate in `PROFILE.md` (in the private `pw-meta` repo),
+projected into `profile-readme/data/profile.json`. Public bio, internship, and
+research copy are synchronized with the 3D site in `tools/sync-profile.py`.
 
 ## The CV
 

content/projects/r1/index.md

@@ -1,9 +1,11 @@
 ---
-title: "LLM for Incentive Design in Mobility Systems"
-summary: "Can an LLM design the incentives that make shared mobility actually work?"
+title: "Scalable Split Learning and Quantization for Vision–Language Models in Medicine"
+summary: "Studying multi-server split learning, low-bit representation transfer, and privacy–utility trade-offs in medical visual question answering."
 date: 2026-01-01
 tags:
   - Research
 ---
 
-An LLM-guided workflow that generates and evaluates interpretable incentive policies for ride-hailing, transit, and shared vehicles — pairing learning-based methods with classical optimization.
+**University of Michigan · 2026–present**
+
+Studying multi-server split learning, low-bit representation transfer, and privacy–utility trade-offs in medical visual question answering.

content/projects/r2/index.md

@@ -1,9 +1,11 @@
 ---
-title: "EPCOT Foundation Model — TSS-to-Expression Data Pipeline"
-summary: "Made a genomic foundation model 2–5× faster to train — and found a 50 kb bug that was quietly wrecking its interpretability."
+title: "Foundation-Model Fine-Tuning for Regulatory Genomics"
+summary: "Fine-tuning EPCOT on single-cell multi-omics data to predict gene expression from chromatin accessibility, with reproducible preprocessing and cell-specific evaluation."
 date: 2026-01-01
 tags:
   - Research
 ---
 
-Pseudobulk pipeline from raw snMultiome (1 bp sparse ATAC → 600 kb HDF5 windows; CP10K-normalized 19,264-gene RNA vectors across 7 cell types). Five bit-equivalent training optimizations for a 2–5× speedup at identical numerics. Reset-head fine-tuning for small-data generalization. Located and fixed a ~50 kb TSS-bin offset.
+**University of Michigan · 2026–present**
+
+Fine-tuning EPCOT on single-cell multi-omics data to predict gene expression from chromatin accessibility, with reproducible preprocessing and cell-specific evaluation.

content/projects/r3/index.md

@@ -1,9 +1,11 @@
 ---
-title: "AI-Guided Environmental Sensing & Adaptive Monitoring"
-summary: "Where should you put the next sensor in Lake Michigan? Let uncertainty decide."
+title: "LLM-Assisted Incentive Design for Mobility Systems"
+summary: "Using structured LLM simulations, bandits, and budget constraints to study data-driven incentives for heterogeneous travelers."
 date: 2026-01-01
 tags:
   - Research
 ---
 
-Reproducible ingestion → reconstruction → uncertainty-quantification workflow for the sparse, seasonal Great Lakes, comparing random / grid / uncertainty-aware sensor-placement rules.
+**University of Michigan · 2026–present**
+
+Using structured LLM simulations, bandits, and budget constraints to study data-driven incentives for heterogeneous travelers.

content/projects/r4/index.md

@@ -1,9 +1,11 @@
 ---
-title: "Applets for Data Analytics in Materials Science"
-summary: "Turning materials-science analysis into something you can just... upload a CSV to."
+title: "AI-Assisted Platform for Scientific Data Analytics"
+summary: "Building a full-stack workflow for materials-science data preparation, interactive model fitting, and human-verified AI assistance."
 date: 2026-01-01
 tags:
   - Research
 ---
 
-JavaScript front end + Python back end serving the materials community: users upload datasets and analyze them with SOTA models and fitting methods, built as reusable FAIR-compliant modules.
+**University of Michigan · 2026–present**
+
+Building a full-stack workflow for materials-science data preparation, interactive model fitting, and human-verified AI assistance.

data/authors/me.yaml

@@ -8,9 +8,9 @@ name:
   given: Tianlang
   family: Chen
   alternate: "陈天朗"
-role: "B.S.E. Candidate, Data Science"
+role: "B.S.E. Candidate · Data Science"
 bio: |
-  I study Data Science at the University of Michigan, Ann Arbor. I came here through the UM-SJTU dual-degree program: the first half of my undergraduate degree was Mechanical Engineering at Shanghai Jiao Tong University's Joint Institute, the second half is Data Science at Michigan. It was less a change of heart than a change of tools.
+  I work across machine learning, scientific data, and full-stack tools for research.
 
 affiliations:
   - name: "University of Michigan, Ann Arbor"
@@ -47,14 +47,12 @@ education:
       Dynamics & Vibrations.
 
 experience:
-  - role: WeCom Full-Stack Developer, Intern
-    org: Mars Asia — Global IT Service Center
+  - role: WeCom Full-Stack Development
+    org: Mars Asia · Global IT Service Center
     start: 2024-06-01
     end: 2025-05-31
     summary: |
-      Full-stack WeCom development in JavaScript, Node.js and Vue.js. API and
-      interface work, data migration, and cross-platform retrieval. Built the
-      automation that kept operations continuous through a system migration.
+      Contributed to internal WeCom tools with JavaScript, Node.js, and Vue.js, supporting APIs, interfaces, data migration, information retrieval, and workflow automation.
 
 skills:
   - name: "Languages"

tools/sync-profile.py

@@ -1,21 +1,22 @@
 #!/usr/bin/env python3
-"""Regenerate this site's content from the single source of truth.
+"""Regenerate this site's profile and synchronized public resume content.
 
 PROFILE.md (in the private pw-meta repo) is the authoritative "who I am". It is
 projected into profile-readme/data/profile.json, and this script projects *that*
 into the two places Hugo reads:
 
     data/authors/me.yaml     the profile: bio, education, internship, skills, ...
-    content/projects/<id>/   one folder per research project
+    content/projects/<id>/   the four Research cards shared with the 3D site
 
 Run it after any change to profile.json, then commit the result. Nothing here is
 hand-maintained; editing me.yaml directly means the next run overwrites you.
 
     tools/sync-profile.py            regenerate
     tools/sync-profile.py --check    exit non-zero if the checked-in files are stale
 
-Cover images (content/projects/*/featured.png) are NOT touched — they are assets,
-and regenerating content must not delete them.
+Identity and skills come from profile.json. The public bio, Mars Internship, and
+four Research descriptions below intentionally match the 3D website. Cover images
+(content/projects/*/featured.png) are not touched.
 """
 
 import argparse
@@ -26,6 +27,44 @@
 HERE = pathlib.Path(__file__).resolve().parent.parent
 SRC = HERE.parent / "profile-readme" / "data" / "profile.json"
 PUBLIC_GITHUB_USERNAME = "jackiectl2"
+PUBLIC_ROLE = "B.S.E. Candidate · Data Science"
+PUBLIC_BIO = (
+    "I work across machine learning, scientific data, and full-stack tools "
+    "for research."
+)
+PUBLIC_INTERNSHIP = {
+    "role": "WeCom Full-Stack Development",
+    "org": "Mars Asia · Global IT Service Center",
+    "start": "2024-06-01",
+    "end": "2025-05-31",
+    "summary": (
+        "Contributed to internal WeCom tools with JavaScript, Node.js, and "
+        "Vue.js, supporting APIs, interfaces, data migration, information "
+        "retrieval, and workflow automation."
+    ),
+}
+PUBLIC_RESEARCH = [
+    {
+        "id": "r1",
+        "title": "Scalable Split Learning and Quantization for Vision–Language Models in Medicine",
+        "summary": "Studying multi-server split learning, low-bit representation transfer, and privacy–utility trade-offs in medical visual question answering.",
+    },
+    {
+        "id": "r2",
+        "title": "Foundation-Model Fine-Tuning for Regulatory Genomics",
+        "summary": "Fine-tuning EPCOT on single-cell multi-omics data to predict gene expression from chromatin accessibility, with reproducible preprocessing and cell-specific evaluation.",
+    },
+    {
+        "id": "r3",
+        "title": "LLM-Assisted Incentive Design for Mobility Systems",
+        "summary": "Using structured LLM simulations, bandits, and budget constraints to study data-driven incentives for heterogeneous travelers.",
+    },
+    {
+        "id": "r4",
+        "title": "AI-Assisted Platform for Scientific Data Analytics",
+        "summary": "Building a full-stack workflow for materials-science data preparation, interactive model fitting, and human-verified AI assistance.",
+    },
+]
 
 
 def q(s: str) -> str:
@@ -57,9 +96,9 @@ def build_author(d: dict) -> str:
   given: Tianlang
   family: Chen
   alternate: {q(ident['name_zh'])}
-role: {q(ident['role'])}
+role: {q(PUBLIC_ROLE)}
 bio: |
-  {d['bio_long'][0]}
+  {PUBLIC_BIO}
 
 affiliations:
   - name: {q(ident['affiliation'])}
@@ -96,14 +135,12 @@ def build_author(d: dict) -> str:
       Dynamics & Vibrations.
 
 experience:
-  - role: WeCom Full-Stack Developer, Intern
-    org: Mars Asia — Global IT Service Center
-    start: 2024-06-01
-    end: 2025-05-31
+  - role: {PUBLIC_INTERNSHIP['role']}
+    org: {PUBLIC_INTERNSHIP['org']}
+    start: {PUBLIC_INTERNSHIP['start']}
+    end: {PUBLIC_INTERNSHIP['end']}
     summary: |
-      Full-stack WeCom development in JavaScript, Node.js and Vue.js. API and
-      interface work, data migration, and cross-platform retrieval. Built the
-      automation that kept operations continuous through a system migration.
+      {PUBLIC_INTERNSHIP['summary']}
 
 skills:
 {skills}
@@ -132,29 +169,20 @@ def build_author(d: dict) -> str:
 
 
 def project_pages(d: dict) -> dict:
-    """slug -> index.md. One per research entry, plus the visible placeholder."""
+    """Return the four public research pages shared with the 3D website."""
     pages = {}
-    for r in d["research"]:
-        pages[r["id"].lower()] = f"""---
+    for r in PUBLIC_RESEARCH:
+        pages[r["id"]] = f"""---
 title: {q(r['title'])}
-summary: {q(r['hook'])}
+summary: {q(r['summary'])}
 date: 2026-01-01
 tags:
   - Research
 ---
 
-{r['detail']}
-"""
-    # The trailing card. Headroom the visitor can see, not a silent gap.
-    pages["more"] = f"""---
-title: {q(d['labels']['research_more'])}
-summary: More work is on the way — this slot exists so the room to grow is visible.
-date: 2025-12-31
-tags:
-  - Research
----
+**University of Michigan · 2026–present**
 
-More projects are in progress.
+{r['summary']}
 """
     return pages