Faculty Profile: Prof. Akari Asai — CMU LTI
Faculty Profile: Prof. Akari Asai — CMU LTI
Position: Assistant Professor（joining Fall 2026） Institution: Carnegie Mellon University, Language Technologies Institute (+ MLD courtesy appointment) Website: https://akariasai.github.io/ Lab: Hiring page: https://akariasai.github.io/groups/ Report date: 2026-06-12
Research Focus
Augmented LMs and agentic systems: RAG, tool use, multi-LM coordination, deep research agents trained with RL. LLM reliability: hallucinations, copyright, unreliable reasoning. AI for science, code, and multilinguality.
Academic Profile
- PhD: University of Washington (NLP, advised by Hannaneh Hajishirzi)
- Research Scientist: Allen Institute for AI (Ai2), 2023–2026
- Visiting student at CMU (2024), collaborated with Graham Neubig
- MIT Technology Review “35 Innovators Under 35” (2025)
- Joining CMU Fall 2026 — lab brand new, PhD slots fully open
Key Publications
| Paper | Venue | Relevance |
|---|---|---|
| Self-RAG: Learning to Retrieve, Generate, and Critique | ICLR’24 | Agentic RAG with self-reflection — foundational RAG + agent work |
| DR Tulu (Deep Research Tulu-8B) | 2025 | First open model trained with RL for long-form deep research; end-to-end open deep research agent |
| A-RAG: Scaling Agentic RAG via… | arXiv Feb 2026 | Scaling agentic RAG systems, multi-step retrieval |
| FlexRAG / FLARE / RQ-RAG | Various | Iterative retrieval, query reformulation, adaptive RAG |
Hiring Status
Actively recruiting Fall 2026 PhD students. Research areas:
- Augmented LMs & agents (RAG, tool use, deep research)
- Safety (hallucinations, copyright, unreliable reasoning)
- AI for science, code & multilinguality
Fit with Weijia Zhang
| Dimension | Assessment |
|---|---|
| Agentic RAG / augmented LMs | ✅ Core focus — Self-RAG, A-RAG |
| RL training for agents | ✅ DR Tulu trained with RL for deep research |
| Agent evaluation | ✅ Building evaluation for retrieval-augmented agents |
| SFT / data pipeline | ✅ Pre-training data curation, instruction tuning |
| GUI / VLM agents | ⚠️ Primarily text-based; some multimodal extension |
| Multi-agent systems | ✅ Multi-LM coordination |
Cold Email Strategy
- Reference: GUIAgentDebugger (self-evolving agent framework, episodic+semantic memory, RAG retrieval) → maps to her Self-RAG + agentic systems
- Reference: OpenManus-RL (RL training for agents) → maps to DR Tulu RL training
- Reference: MSRA TextAnalysisSFT (SFT data pipeline) → data engineering experience
- Mention: Excited about combining RAG + RL for more reliable long-horizon agents
Verdict
P1 套磁，最高优先级。 Joining Fall 2026 with a brand-new lab. Research perfectly matches Weijia’s work. Officially recruiting PhD students. Should email before the Fall 2026 application deadline (Dec 2026 for Fall 2027 PhD, or reach out now to express interest for future collaboration).