About

[Latest update: Apr, 2026]

Hi! My name is Arash Ashrafzadeh, and I'm an AI researcher and engineer, and
I am currently on the job market!
I'm open to NLP researcher, data scientist and AI engineer roles in industry and academia. Ideally, I would like to study small models of real-time dialogue understanding, meaning that my interests are at the intersection of:

• Mechanistic Interpretability → understanding how incremental processing works in (recurrent) language models
• Neuro-Symbolic AI → building cognitively-plausible (data efficient, generalisable and incremental) models of language
• Edge AI → optimised model deployment, on-device learning and efficient inference with smol/quantised models
• Voice AI → benchmarking end-to-end vs modular approaches for robust real-time conversational AI

If you are interested in collaborating with me, please feel free to contact me by email on incrementaliser AT gmail.com.

---

Education

I recently finished my PhD in NLP at Heriot-Watt university. I was supervised by Dr. Arash Eshghi and Dr. Julian Hough. Before this, I received my BSc in Computer Science from Amirkabir University of Technology (Tehran Polytechnic).
Broadly speaking, my research topic has been Computational Linguistics for Conversational AI, and more specifically, I have been working on Incremental Grounded Language Learning. In particular, I have been interested in Incremental Semantic Parsing and Incremental NLG from Symbolic Semantic Representations. If you are curious to know more, I have made a poster for it here 👀!

---

Publications & Projects

• Learning to generate and corr- uh I mean repair language in real-time. [arxiv]
  Arash Eshghi, Arash Ashrafzadeh. SemDial 2023.

• Persian Question Answering system using TF-IDF.
  Arash Ashrafzadeh. BSc Project (2019). Amirkabir University of Technology.

---

News

---

Blog Posts