Heydar Soudani

Heydar Soudani

PhD Candidate
Radboud University
Nijmegen, The Netherlands

I am a final-year PhD candidate at Radboud University, where I focus on developing trustworthy AI systems. My research centers on retrieval-augmented generation (RAG), deep research agents, and uncertainty quantification, with the goal of building reliable AI systems capable of solving complex information-seeking and reasoning tasks.

My work has been published in leading academic venues, including ACL, EMNLP, SIGIR, ACM Computing Surveys, COLING, SIGIR-AP, and ICTIR.

Alongside my academic research, I actively collaborate with industry as an Applied AI and Data Science Researcher. Most recently, I completed a research internship at Thomson Reuters, where I developed agentic search solutions for the legal domain. Prior to that, I worked with KPN to integrate uncertainty quantification methods into their RAG systems, improving the reliability and trustworthiness of AI-generated responses.

Education
Feb 2023 — Present

Ph.D. Computer Science

Radboud University, Netherlands

Data Augmentation for Conversational AI

Supervisor: Faegheh Hasibi

Sep 2019 — Jun 2022

M.Sc. Computer Engineering

Sharif University, Iran

Predicting Novelty Concepts in Data Stream

Supervisor: Hamid Beigy

Sep 2014 — Sep 2019

B.Sc. Electrical Engineering

Polytechnic of Tehran, Iran

Providing new cloud services on the Kubernetes platform

Supervisor: Hassan Taheri

Sep 2016 — Sep 2018

B.Sc. Computer Engineering

Polytechnic of Tehran, Iran

Minor degree, taken alongside the Electrical Engineering major.

Work Experience
Feb 2023 — Present

PhD Candidate

Radboud University, Nijmegen, Netherlands

Trustworthy Generative Information Access. Researching deep research agents and uncertainty quantification.

Published at ACL, SIGIR, ACM Computing Surveys, SIGIR-AP, and ICTIR.

Supervisors: Faegheh Hasibi, Arjen de Vries

Information RetrievalAgentic SystemsUncertainty QuantificationRL Post-training
Nov 2025 — May 2026

Data Scientist Intern

Thomson Reuters · Zug, Switzerland · Remote

Researched and developed agentic search systems for the legal domain. Published at EMNLP.

Supervisors: Navid Rekabsaz, Elizabeth Lingg

R&DAgentic SearchLegal DomainAWS
May 2025 — Oct 2025

Data Scientist Intern

KPN · Amsterdam, North Holland, Netherlands · On-site

Applied uncertainty quantification methods to internal RAG systems to make their responses more reliable.

Supervisor: Gianluigi Bardelloni

R&DRAGLess-popular Domain
Dec 2017 — Jun 2022 · 4 yrs 7 mos

Web Developer

Andishe Fartak Amirkabir (Atrovan) · Part-time · Tehran Province, Iran

Developed web applications for:

  • Smart metering system (AtroMeter)
  • Fleet management system (Navgoon)
  • Home automation (Smart Home)
  • Building management system (BMS)
Web DevelopmentReact.jsJavaScript
Jan 2016 — Oct 2017

Research Assistant

Control of Multi Vehicle Systems Lab (CMVL) · Polytechnic of Tehran

Collaborated on AVR and Raspberry Pi programming, computer vision for localization, and PCB design with Altium Designer.

Supervisor: Farzaneh Abdollahi

Computer VisionEmbedded SystemsRobotics
Technical Skills

⬡ Languages

Python C / C++ SQL JavaScript LaTeX

⬡ Infrastructure & Tools

Git Linux HPC / Slurm Docker Kubernetes AWS vLLM Weights & Biases

⬡ ML / LLM

PyTorch TensorFlow Hugging Face Transformers LLM Fine-tuning (LoRA / PEFT) RL Post-training LLM Evaluation & Benchmarking Uncertainty Quantification Prompt Engineering NumPy Pandas scikit-learn

⬡ RAG & Agents

Retrieval-Augmented Generation (RAG) AI Agents / Agentic Search LangChain / LangGraph Dense & Sparse Retrieval Reranking FAISS OpenAI / LLM APIs
Publications
2026

When Deep Research Agents Stagnate: Enhancing Reasoning with Retrieval-Aware Agent Control

Heydar Soudani, Elizabeth Lingg, Faegheh Hasibi, Navid Rekabsaz
Conference on Empirical Methods in Natural Language Processing (EMNLP)
2026

Uncertainty Quantification for Retrieval-Augmented Reasoning

Heydar Soudani, Hamed Zamani, Faegheh Hasibi
The 49th International ACM SIGIR Conference on Research and Development in Information Retrieval
2026

Total Recall QA: A Verifiable Evaluation Suite for Deep Research Agents

Mahta Rafiee, Heydar Soudani, Zahra Abbasiantaeb, Mohammad Aliannejadi, Faegheh Hasibi, Hamed Zamani
The 49th International ACM SIGIR Conference on Research and Development in Information Retrieval
2026

Uncertainty Quantification for Multimodal Retrieval Augmented Generation

Simon Binz, Heydar Soudani, Faegheh Hasibi
The 2026 International ACM SIGIR Conference on Innovative Concepts and Theories in Information Retrieval (ICTIR)
2026

A Survey on Recent Advances in Conversational Data Generation

Heydar Soudani, Roxana Petcu, Evangelos Kanoulas, Faegheh Hasibi
ACM Computing Surveys
2025

Why Uncertainty Estimation Methods Fall Short in RAG: An Axiomatic Analysis

Heydar Soudani, Evangelos Kanoulas, Faegheh Hasibi
Findings of the Association for Computational Linguistics (ACL)
2024

Fine Tuning vs. Retrieval Augmented Generation for Less Popular Knowledge

Heydar Soudani, Evangelos Kanoulas, Faegheh Hasibi
The 2024 Annual International ACM SIGIR Conference on Research and Development in Information Retrieval in the Asia Pacific Region (SIGIR-AP)
2022

Persian Natural Language Inference: A Meta-learning Approach

Heydar Soudani, Mohammad Hassan Mojab, Hamid Beigy
International Conference on Computational Linguistics (COLING)
Paper Slides Code Poster
Academic Service
Tutorials, Talks & Organizing
  • Workshop DIR 2025
    The 22nd Dutch-Belgian Information Retrieval Workshop
  • Doctoral consortium SIGIR 2025
    Enhancing Knowledge Injection in Large Language Models for Efficient and Trustworthy Responses
  • Tutorial WebConf 2024
    Data Augmentation for Conversational AI (second edition)
  • Tutorial CIKM 2023
    Data Augmentation for Conversational AI (first edition)
Reviewing
Full papers
  • ACL 2025
  • EMNLP 2026
  • SIGIR 2026
  • CIKM 2026
Short papers
  • SIGIR 2026
  • ECIR 2026
  • WSDM 2026
Journals
  • TOIS
  • ACM Computing Surveys
Thesis Supervision
  • Uncertainty Quantification for Multimodal Retrieval-Augmented Generation
  • Failure Modes of Deep Research Agents in Context-Aided Forecasting