Faculty Profile: Prof. Dilek Hakkani-Tür — UIUC ConvAI Lab
Faculty Profile: Prof. Dilek Hakkani-Tür — UIUC ConvAI Lab
Position: Full Professor (joined UIUC 2024) Institution: UIUC, Siebel School of Computing and Data Science Lab: ConvAI Lab @ UIUC (co-founded with Prof. Gokhan Tur, 2024) Website: https://uiuc-conversational-ai-lab.github.io/ Report date: 2026-06-12
Research Focus
Conversational agents, large language models, task-oriented dialogue systems, LLM reasoning, evaluation frameworks, persuasion. Building open-source agentic LLMs (CoALM series).
Academic Profile
- h-index: 77 (Google Scholar)
- 400+ publications, 100+ granted patents
- 20+ years industry research:
- Amazon Alexa AI, Senior Principal Scientist (2018–2023)
- Google Research, Dialogue Research Lead (2016–2018)
- Microsoft Research, Principal Researcher (2010–2016)
- 2026 Grainger College Dean’s Award for Excellence in Research
- Funding: DARPA, IARPA, Department of Education, Amazon, Microsoft
Key Publications (2025–2026)
- CoALM (2025): Unified Conversational Agentic Language Model — CoALM-405B is the largest fully open-source conversational agent LLM, ranked #1 on Berkeley Function Calling Leaderboard V3
- ReIn (ICLR 2026): Conversational error recovery with reasoning inception
- ATOD (2026): Evaluation framework and benchmark for agentic task-oriented dialogue
- DialDefer (2026): Detecting and mitigating LLM dialogic deference
- Current Agents Fail to Leverage World Model as Tool for Foresight (2026)
- TT-SI (ACL 2026): Self-Improving LLM Agents with Test-Time Training
Lab Snapshot
- 10 PhD students, 3 MSc students (as of 2024–2025)
- Lab founded 2024 — no PhD graduates yet, no placement data
Fit with Weijia Zhang
| Dimension | Assessment |
|---|---|
| LLM agents / agentic AI | ✅ Core focus (CoALM, task-oriented agents) |
| Post-training (SFT / RL) | ✅ SFT for dialogue agents, self-improving agents |
| Evaluation / benchmarks | ✅ Strong (ATOD, error recovery) |
| Reasoning | ✅ Reasoning in dialogue (ReIn, world model) |
| VLM / GUI agents | ❌ Primarily text-based dialogue |
| RAG / retrieval | ⚠️ Adjacent via retrieval in dialogue |
Risks
- No alumni placement data — lab founded 2024, zero PhD graduates
- No verified frontier placements (OpenAI / Anthropic / DeepMind) — too early to evaluate
- Domain: dialogue systems — overlaps with agentic AI but not GUI or multimodal
- Amazon pipeline bias — Amazon is primary industry funder
Connection to Weijia
- Direct research overlap: agentic LLMs, SFT for agents, evaluation
- Industry-to-academia transition: deep industry network at Amazon/Google/Microsoft → strong internship pathways
- Potential route: email Hakkani-Tür citing GUIAgentDebugger + OpenManus-RL, which overlaps with her self-improving agents work
Verdict
Strong research match (agentic LLMs, SFT, evaluation frameworks, self-improving agents). Significant unknown: no placement record from UIUC lab. Recommended as PhD second choice or co-advisor alongside Heng Ji.