Advisor Dossier: Prof. Jiaxuan You — UIUC (Siebel School of Computing and Data Science)

Student: Weijia Zhang | M.S. CS, Yale University (Aug 2026 – May 2028, Thesis Track, Full Scholarship) Assumed goals (inferred from CV): Industry-research 60% · Academia (PhD after M.S.) 40% Report date: 2026-06-11

⚠️ Critical Institutional Mismatch — Read First

Jiaxuan You is at UIUC (Champaign, IL). Weijia Zhang is an incoming M.S. student at Yale University.

This creates a structural incompatibility: You cannot serve as Weijia’s primary M.S. thesis advisor at Yale. This dossier therefore evaluates two distinct scenarios:

• Scenario A (primary evaluation): Weijia applies to Jiaxuan You’s PhD program at UIUC — either abandoning the Yale M.S. offer or completing it and then applying to UIUC PhD.
• Scenario B (secondary evaluation): Jiaxuan You as an external research collaborator, remote mentor, or future PhD supervisor after the Yale M.S.

Recommendation upfront: Scenario B is likely the better near-term strategy. Weijia is already committed to Yale M.S. (August 2026). Jiaxuan You is the most research-aligned advisor of all three evaluated — establishing a remote collaboration or paper co-authorship during Yale M.S. positions Weijia well for a future UIUC PhD application or UIUC-connected industry placement. Do not discard this advisor simply because of institutional distance.

1. Executive Summary

Top critical risks/unknowns:

1. Zero alumni data from UIUC lab (structural): Jiaxuan You joined UIUC as faculty in Fall 2024 — only ~2 years into his tenure-track position. No PhD students have graduated yet. Placement track record is completely empty.
2. Frontier lab placement: zero confirmed — No UIUC students from this lab have confirmed internships or full-time roles at OpenAI, Anthropic, DeepMind, Meta FAIR, or MSR.
3. Institutional mismatch: Jiaxuan You is at UIUC; Weijia is at Yale. Cannot be primary M.S. thesis advisor. Any collaboration must be remote or post-M.S.
4. Lab funding unconfirmed: No NSF CAREER or confirmed federal grants found. The lab is very new; grant pipeline likely in early stages.

One-line verdict: Jiaxuan You’s research direction is the strongest match of all three advisors evaluated — but the institutional mismatch and zero-alumni track record make him a “future PhD advisor” target rather than a viable Yale M.S. thesis advisor right now.

Strongest pros:

• Research overlap with Weijia is exceptional: MultiAgentBench, LLM agent infrastructure, ResearchTown, GraphPlanner, memory-augmented agents — this is almost exactly what Weijia has been building
• Lab name (“U Lab”) and stated focus: “LLM Agent infrastructure and application” — verbatim match to Weijia’s MSRA and OpenManus-RL work
• Jiaxuan You worked at NVIDIA Research as Sr. Research Scientist (2022–2024) — direct industry research pathway, not just an academic network
• Leskovec → Kumo.ai → NVIDIA (~$400M acquisition, June 2026) [1]: You has direct access to NVIDIA Research, now the single largest AI infrastructure company
• He collaborates with Xiangru Tang (Cohan’s Yale PhD student), creating a Yale–UIUC bridge that Weijia could traverse from Yale
• UIUC is Weijia’s undergraduate home — strong informal network advantage

Strongest cons:

• Cannot advise Yale M.S. thesis — institutional incompatibility
• Only 2 years as faculty — possibly the youngest lab of the three, zero graduation data
• No confirmed student internship pipeline at frontier labs
• UIUC Champaign is a less vibrant tech hub than the coastal cities; industry recruiting requires more self-initiative
• Very early career (joined at ~26 years old) — mentorship systems, advising norms, and grant structures are not yet established

Score snapshots:

• Four-Dimension Fit Score: 63/100 → Proceed with caution (base)
• AI Industry Outcome (industry-research track): 51/100 → Proceed with caution (no additional cap)
• Coverage gate: Low (zero alumni rows) + weak frontier evidence → caps at Significant concerns

Coverage: Low — only one confirmed UIUC student (Tao Feng), zero graduates, zero alumni placement data. Two of three research agents hit API rate limits; student roster data is partially unavailable.

Concrete next steps:

1. Reach out to Jiaxuan You NOW (before arriving at Yale): Email him about your GUIAgentDebugger work and OpenManus-RL contribution. His MultiAgentBench directly targets the same evaluation gap. This is a warm email write, not a cold one.
2. Connect through Xiangru Tang (Cohan’s student at Yale): Tang co-authored MultiAgentBench with Jiaxuan You — this is a warm 1-hop intro channel once Weijia arrives at Yale in August.
3. Target a joint paper: A natural project is extending GUIAgentDebugger’s error taxonomy to multi-agent coordination failures (bridging You’s MultiAgentBench + Weijia’s taxonomy work). Pitchable to both as first-author work.
4. Use him as a PhD target post-M.S.: Complete Yale M.S. with Cohan, build Jiaxuan You relationship through collaboration, apply to UIUC PhD with strong Yale publication backing.
5. Verify UIUC PhD funding model: Ask directly about RA funding, TA requirements, and whether PhD students from his lab get GPU time for RL-scale training.

2. Critical Problems First

3. Strong Pros and Strong Cons

Pros

• Best research overlap of all three advisors: Jiaxuan You’s lab is explicitly focused on “LLM Agent infrastructure and application.” His 2024–2026 papers (MultiAgentBench, ResearchTown, GraphPlanner, Thought-Retriever, GraphRouter) map almost precisely onto Weijia’s project portfolio (GUIAgentDebugger, OpenManus-RL, MSRA Excel Copilot agent pipeline).
• Direct NVIDIA pathway: You worked at NVIDIA as Senior Research Scientist (2022–2024) before joining UIUC [2]. This is direct industry research placement evidence at a top-tier lab — not a theoretical network connection.
• Kumo.ai → NVIDIA acquisition (~$400M, June 2026): Jure Leskovec (You’s PhD advisor) founded Kumo.ai; NVIDIA acquired it [1]. You has now connected his two career worlds: Stanford-graph-ML and NVIDIA. This makes the NVIDIA pathway even more direct.
• OpenManus-RL is a warm email: OpenManus-RL has 60K+ GitHub stars and is directly in Jiaxuan You’s domain. He almost certainly knows the project. Weijia is a core contributor. This makes cold outreach warm.
• MultiAgentBench co-authored with Xiangru Tang (Cohan’s Yale student) [3]: There is already a live Yale-UIUC collaboration bridge that Weijia can use.
• UIUC is Weijia’s undergraduate home: Familiar academic culture, informal network of peers, faculty awareness of UIUC CS students. UIUC’s C.W. Gear Award (which Weijia won) is known to UIUC faculty.
• Leskovec network (now NVIDIA-centric): Beyond just graph ML, the Leskovec → NVIDIA network now covers the most powerful AI infrastructure company.
• Teaching CS 598 JY2: Topics in LLM Agents [4]: This course signals that You is actively building a PhD student community around exactly Weijia’s interest area.

Cons

• Cannot advise Yale M.S. — period. All other evaluation is contingent on either a remote collaboration or a future PhD.
• 2 years as UIUC faculty — possibly the least established lab of all three advisors. He is ~26-28 years old; systems, expectations, and mentorship norms are still being built.
• No PhD student graduation precedent at UIUC. The only evidence of advising capability comes from his Stanford days (as a PhD student himself, not a primary advisor).
• UIUC Champaign location: Less frequent visits from frontier lab recruiters than Stanford/MIT/CMU/Yale. Students often have to be more proactive about internship networking.
• Lab funding is uncertain: Very early career, likely submitting NSF and DARPA proposals. Stipend security over a 5-year PhD is less certain than at a more established lab.
• Graph ML baggage: Despite the clear pivot, some of You’s students may still be working on core GNN problems (Tao Feng has GraphRouter and GraphPlanner — graph-focused even if LLM-integrated). This may not be the pure LLM-agent lab that Weijia envisions.

4. Academic Profile

Position: Assistant Professor, Siebel School of Computing and Data Science, UIUC (joined Fall 2024) [5]. Previously: Senior Research Scientist, NVIDIA (2022–2024) [2]; Lecturer (brief), Stanford CS.

PhD: Stanford University, Computer Science, 2017–2022, supervised by Jure Leskovec (SNAP Lab) [6]. PhD received ~age 24.

Undergrad: Tsinghua University, B.E. Automation + B.S. Economics (dual degree) [6].

Lab: U Lab at UIUC. GitHub: github.com/ulab-uiuc. Stated focus: “LLM Agent infrastructure and application” [7].

Citations: ~24,037 Google Scholar (as of ~2026, from search snippets) [8]. Semantic Scholar: ~1,471 highly influential citations; ~31 indexed papers [9]. Bulk of citations come from Stanford-era foundational graph ML papers.

Landmark papers:

• GraphRNN (ICML 2018): generative model for graphs [10]
• GCPN (NeurIPS 2018): RL + GNN for molecular graph generation [11]
• P-GNN (ICML 2019 oral): position-aware GNNs [12]
• Design Space for GNNs / GraphGym (NeurIPS 2020 spotlight) [13]
• ID-GNN (AAAI 2021): identity-aware GNNs [14]

Recent pivot papers (2024–2026):

• RelBench (NeurIPS 2024 D&B): relational deep learning benchmark [15]
• The Virtual Lab (Nature, 2024): LLM multi-agent for nanobody design [16]
• MultiAgentBench (ACL 2025 main): evaluating LLM agent collaboration and competition [3]
• LLM-Based Multi-Agent Systems as Scalable Graph Generative Models (ACL 2025 Findings, with Rex Ying) [17]
• ResearchTown (ICML 2025): simulator of human research community [18]
• GraphRouter (ICLR 2025): graph-based LLM routing [19]
• GraphPlanner (ICLR 2026): graph memory-augmented multi-agent routing [20]
• Thought-Retriever (TMLR 2026): memory-augmented agentic retrieval [21]

Awards: J.P. Morgan AI PhD Fellowship (2021) [22]; ATLAS/Laude Institute Moonshots Honorable Mention [23].

Industry connections: NVIDIA (direct employment 2022–2024); Kumo.ai (Leskovec startup, acquired by NVIDIA ~$400M June 2026 [1]); Pinterest (Stanford-era connection); Kumo adjacent: Weihua Hu, Yiwen Yuan.

5. Alumni Outcomes and Graduation Windows

PhD Graduates from UIUC (n=0)

No PhD students have graduated from Jiaxuan You’s UIUC lab. He joined Fall 2024.

Current PhD Students (Partially Identified)

Coverage note: Due to API rate limits on agents 1 and 2, the full student roster could not be confirmed. Tao Feng is the only student with a complete, independently verifiable co-author trail. Other students listed are inferred from paper co-authorship and may include external collaborators.

External Collaborators (Not UIUC Students)

• Zhenhailong Wang: MultiAgentBench co-author; UIUC CS PhD but advised by Heng Ji, not You
• Xiangru Tang: MultiAgentBench co-author; Yale CS PhD under Arman Cohan [3]
• Weihua Hu: Stanford/Kumo.ai-affiliated; RelBench co-author
• Rex Ying: Yale faculty; recurring co-author; ACL 2025 [17]

6. Placement Distribution and Attrition Analysis

PhD cohort (n=0): Cannot be analyzed. Lab is 2 years old.

Tao Feng (only identifiable active student): 4 confirmed top-venue papers (ICLR 2025, ICLR 2026, 2× TMLR) in ~2 years of PhD — exceptional output rate. He is the best current signal of what the lab can produce.

Distribution summary:

• Upper tail: Unknown (no graduates)
• Median: Unknown
• Lower tail: Unknown
• Frontier readiness: No evidence — uncharted

Attrition: No evidence of non-completions found. Lab is too new to observe attrition patterns.

NVIDIA pathway assessment: Jiaxuan You’s direct NVIDIA path is the most concrete industry research channel for this lab. NVIDIA Research (including teams that worked with Kumo.ai) is a plausible destination for students working on graph-augmented LLM agents. This is not documented but is structurally possible.

7. Data Coverage Dashboard

Overall coverage confidence: Low (structural — zero alumni rows to analyze; also 2/3 research agents hit rate limits)

Coverage gate result: Low coverage + zero frontier evidence → caps final verdict at Significant concerns ✓

What missing data would most change the verdict:

1. Tao Feng’s post-PhD placement (expected ~2027–2028) — this will be the lab’s first real data point
2. Any confirmed internship from any UIUC student at NVIDIA Research, OpenAI, or Anthropic
3. Jiaxuan You’s confirmed NSF/DARPA grants
4. Full current student roster (2 agents hit rate limits; roster may be larger than identified)

8. Four-Dimension Risk and Fit Assessment

Goal weights (blended: 60% industry-research + 40% academia): Survival: 27 | Academic: 27 | Industry: 31 | Happiness: 15

Four-Dimension Fit Score: 62 × 0.27 + 65 × 0.27 + 58 × 0.31 + 70 × 0.15 = 16.74 + 17.55 + 17.98 + 10.50 = **62.8/100**

Base verdict: Proceed with caution (50–74 range)

9. AI Industry Outcome Scorecard (Industry-Research Track)

Industry-research track total: 51/100 → Proceed with caution (50–74 range; no scorecard-level cap)

Verified Frontier Placement Table

Frontier pipeline funnel: No data. All stages (Applied → Interviewed → Interned → Return offer → Full-time) are unknown for UIUC lab students.

Frontier readiness: Limited (0 confirmed full-time, 0 confirmed internships from lab students)

10. Verdict and Score Reconciliation

Final verdict: ⚠️ Significant concerns (62.8/100 base, capped by coverage gate)

Critical context: The “Significant concerns” verdict here is driven almost entirely by structural factors (lab too new, zero data) rather than quality signals. The underlying four-dimension score (62.8) and industry-research track (51) are actually comparable to or better than Rex Ying. The gap from Cohan (71) comes from the lack of any measurable alumni outcomes — not from evidence of poor placement.

Institutional scenario note:

• PhD at UIUC under Jiaxuan You: Verdict = Significant concerns (coverage gate). Upside is high; data to support it doesn’t exist yet.
• Remote mentor/future PhD pathway from Yale M.S.: No formal verdict possible (not thesis advising). Strategically, this is the best risk-adjusted path — build the relationship without taking the structural risk of the zero-alumni lab.

11. Personalized Fit (Weijia Zhang × Jiaxuan You)

Research Overlap

Most Promising Intersection Projects

1. Multi-agent failure mode taxonomy and benchmark: Weijia’s GUIAgentDebugger has a 4-category, 29-subtype taxonomy of individual agent failures. Jiaxuan You’s MultiAgentBench focuses on coordination failures in multi-agent systems. Combining these: a comprehensive failure taxonomy for multi-agent LLM systems that covers both individual and coordination failures. Natural joint paper (Weijia first author, You/Cohan as senior authors if at Yale).

2. Graph-augmented memory for agent debugging and self-improvement: Weijia’s GUIAgentDebugger uses dual-layer memory + intent-aware RAG to learn from failure trajectories. You’s GraphPlanner/Thought-Retriever use graph-structured memory for agent routing. Intersection: building a graph-structured failure memory that enables agents to generalize from past errors across semantically related tasks.

3. RL training for multi-agent coordination in structured tasks: Weijia has VERL + step-level reward design experience from OpenManus-RL. You has GCPN (RL+graph) precedent and is pivoting toward agentic systems. Intersection: applying RL training (Weijia’s skill) to the multi-agent coordination problem (You’s current focus).

Key Friction Points

1. Institutional distance: Remote collaboration is less productive than in-person. Being in different time zones is not the case (both Chicago/CT area), but being on different campuses limits lab immersion.
2. Graph ML entry cost: Even though You’s lab has pivoted toward agents, the underlying graph motif (GraphRouter, GraphPlanner) still requires some graph ML background that Weijia doesn’t have.
3. Very young PI: Jiaxuan You is ~28 years old with 2 years of faculty experience. Advising style, feedback speed, and student management systems are not yet proven. The quality of the PI-student relationship at this stage is highly variable.
4. Competition for attention: Tao Feng appears to be You’s primary student and has 4 top-venue papers already. Weijia (as a remote collaborator or new PhD student) may find bandwidth constrained.

Weijia’s Unique Edge

Jiaxuan You almost certainly knows OpenManus/OpenManus-RL (60K+ stars in exactly his research space). Weijia’s core contributor status on that project is a direct warm email credential — not just a resume item.

Network Complementarity

• Jiaxuan You’s network: NVIDIA, Leskovec/Stanford, Kumo.ai, graph ML industry (Meta AI Graphs, Pinterest, LinkedIn)
• Weijia’s existing network: MSRA (Microsoft), Tencent WeChat, OpenManus/ByteDance adjacent
• Together: complementary US+China industry research access

One-line fit verdict

The research overlap between Weijia and Jiaxuan You is the best of all three advisors — but institutional distance and a 2-year-old lab with zero placement data make the case for a “build relationship from Yale, apply to PhD later” strategy rather than betting everything on a very young unknown.

12. Alumni Impact and Connection Mapping (Prioritized)

Key insight: Xiangru Tang is at Yale and is a MultiAgentBench co-author with Jiaxuan You. Weijia should treat Tang as the primary introduction pathway to You — not cold email.

13. Funding and Resources

Compute risk: The lab focuses on LLM agents, which requires GPU time. NVIDIA connection may provide access. But RL-at-scale training (Weijia’s specialty) may still require significant compute resources not guaranteed at a 2-year-old lab.

14. Research Gaps (What We Don’t Know)

1. Full student roster: 2/3 research agents hit API rate limits; complete student roster not available. Tao Feng confirmed; others (Kunlun Zhu, Zijie Lei, Peixuan Han, Haozhen Zhang) are inferred from co-authorship and may include external collaborators.
2. Grant status: No confirmed NSF/DARPA/NIH grant numbers, amounts, or timelines.
3. Tao Feng’s post-PhD plans: He’s ~year 2 with 4 papers; expected graduation ~2027. His destination will be the lab’s first real placement signal.
4. Jiaxuan You’s advising style: Too new; no ex-student testimony available. All evidence is collaborative, not supervisory.
5. Relationship with Kumo.ai/NVIDIA post-acquisition: Does You have formal research collaboration with NVIDIA? Can students access NVIDIA Research resources?
6. M.S. advising policy: Can UIUC admit M.S. thesis students in addition to PhDs? What does Jiaxuan You’s M.S. advising model look like?

15. Questions to Ask

Questions for Prof. Jiaxuan You

1. What are the major open problems in LLM agent infrastructure, and where does your lab have unique positioning vs. Stanford/MIT/CMU?
2. If I’m starting at Yale M.S. (not your PhD program), is there a realistic path to remote collaboration on a joint paper? What would that look like?
3. What is your NVIDIA connection like post-Kumo acquisition — can students in your lab access NVIDIA Research collaborations or referrals?
4. What is your internship policy? Do students typically intern at AI companies during their PhD?
5. What is the funding model for PhD students — RA, TA, or fellowship? What happens if a grant isn’t renewed?
6. Tao Feng has 4 papers in 2 years — is that typical of your lab, or is he exceptional?
7. Are you planning to admit any students from outside UIUC as visiting researchers or remote collaborators?
8. What would a typical PhD project look like for someone with my background in agent RL and evaluation?

Questions for Current Students (Tao Feng)

1. How does Jiaxuan advise in practice — weekly 1:1s? Ad hoc? How fast does he review drafts?
2. How much of the research direction is Jiaxuan-driven vs. student-driven?
3. What is the compute situation — do students have GPU access for RL-scale experiments?
4. Is the lab culture collaborative or individually focused?
5. Has Jiaxuan helped with internship applications? Does he make calls for students?
6. What do you wish you knew before joining?

High-Uncertainty, High-Impact Verification Questions

1. Ask Xiangru Tang (Yale, in person from August 2026): “What is Jiaxuan You like to collaborate with? Did he drive the project, review frequently, and support your authorship?”
2. Ask Jiaxuan You directly: “Have any of your UIUC students interned or applied to OpenAI, Anthropic, or Google DeepMind yet?”
3. Search UIUC CS grad admissions profile: Does Jiaxuan You appear on recent admitted PhD student advisor matches in fall 2024 or 2025?
4. Check NSF Award Search for Jiaxuan You — any grants awarded or pending?

16. Strategic Recommendation: Optimal Path for Weijia

Given the institutional mismatch and the exceptional research fit, the optimal strategy is not to choose between Jiaxuan You and the Yale advisors — it’s to use both.

Recommended 2-Phase Plan

Phase 1: Yale M.S. (2026–2028) — Build relationship with Jiaxuan You from Yale

• Primary thesis advisor: Arman Cohan (better frontier lab network, can advise at Yale)
• External collaborator/mentor: Jiaxuan You
• Entry point: Contact Xiangru Tang immediately upon arriving at Yale (he’s in Cohan’s lab and co-authored with You); get a warm intro to You
• Target a joint paper: GUIAgentDebugger × MultiAgentBench = comprehensive multi-agent failure taxonomy. Pitch to both Cohan and You as a natural collaboration.
• By end of M.S.: have 1–2 publications, with You knowing Weijia’s work quality firsthand

Phase 2: UIUC PhD application (2027–2028)

• Apply to UIUC CS PhD under Jiaxuan You, backed by:
  • Yale M.S. thesis from Cohan (strong Yale recommendation)
  • Coauthored paper with You (direct evidence of ability to contribute)
  • OpenManus-RL contributor status (warm email credential)
  • Cohan’s recommendation (cross-institution validation)
• By 2028, You’s lab will have 4 years of students — Tao Feng’s placement will be public

Why this is better than either choosing alone:

• Avoids the zero-data risk of betting entirely on a 2-year-old lab for a 2-year M.S.
• Keeps the Yale M.S. scholarship secured
• Builds a real publication relationship with You before committing to a 5-year PhD
• Gives Weijia two network channels: Cohan’s (Meta FAIR, Anthropic, Google) + You’s (NVIDIA, Leskovec/Stanford)

Sources