Mohammad Aghajani Asl

Tehran, Iran

Open to PhD opportunities

I am an AI/NLP researcher and engineer working on trustworthy, evidence-grounded language-model systems. My research focuses on agentic retrieval-augmented generation, multi-hop reasoning, automated fact-checking, and hallucination mitigation, with particular interest in Persian and other low-resource settings.

I developed FAIR-RAG, an evidence-driven iterative RAG framework that evaluates evidence sufficiency, identifies missing information, and adaptively refines retrieval before generating the final answer. My recent work also includes IslamicPCQA, introduced through an IEEE TASLP publication on Persian complex question answering, and FARSIQA, an end-to-end agentic QA system designed for a sensitive, knowledge-intensive domain.

I hold an M.Sc. in Physics – Complex Systems and a B.Sc. in Physics from Sharif University of Technology. My quantitative background shapes how I approach AI research: through rigorous evaluation, explicit modeling of uncertainty, and close attention to the mechanisms behind a system’s behavior rather than only its final outputs.

Alongside academic research, I have designed and implemented applied AI systems for evidence-grounded scholarly writing, fraud-risk analysis, large-scale Persian NLP, information retrieval, and production data infrastructure.

Research Interests

Trustworthy and Faithful AI: automated fact-checking, claim–evidence verification, hallucination mitigation, and evidence-grounded generation
Agentic RAG and Multi-hop Reasoning: adaptive retrieval, query refinement, evidence assessment, and multi-agent architectures
Information Retrieval and Question Answering: semantic search, hybrid retrieval, dataset construction, and RAG evaluation
Persian and Low-resource NLP: multilingual models, Persian datasets, domain adaptation, and reliable AI for specialized domains
AI for Science: language-model systems that support structured, verifiable scientific and scholarly research

Current Focus

I am currently seeking PhD opportunities in NLP, Trustworthy AI, Agentic AI, Automated Fact-Checking, and AI for Science. I am particularly interested in research on reliable language models that can reason over multiple sources, recognize missing evidence, and produce transparent, verifiable outputs.

For research collaborations, PhD opportunities, or applied AI projects, the best way to reach me is by email.

selected publications

IEEE TASLP
IslamicPCQA: A Dataset for Persian Multi-hop Complex Question Answering in Islamic Text Resources

Arash Ghafouri, Mohammad Aghajani Asl, Hasan Naderi, and Mahdi Firouzmandi

IEEE Transactions on Audio, Speech, and Language Processing, Jul 2025

Abs DOI Bib HTML

IslamicPCQA is a Persian multi-hop complex question-answering dataset containing 12,282 manually constructed question-answer pairs derived from nine Persian encyclopedias. The dataset includes sentence-level supporting facts and benchmark evaluations using Persian and multilingual language models.
@article{ghafouri2025islamicpcqa, title = {IslamicPCQA: A Dataset for Persian Multi-hop Complex Question Answering in Islamic Text Resources}, author = {Ghafouri, Arash and Aghajani Asl, Mohammad and Naderi, Hasan and Firouzmandi, Mahdi}, journal = {IEEE Transactions on Audio, Speech, and Language Processing}, volume = {33}, pages = {3801--3812}, year = {2025}, month = jul, publisher = {IEEE}, doi = {10.1109/TASLPRO.2025.3587450}, }
arXiv
FAIR-RAG: Faithful Adaptive Iterative Refinement for Retrieval-Augmented Generation

Mohammad Aghajani Asl, Majid Asgari-Bidhendi, and Behrooz Minaei-Bidgoli

arXiv preprint arXiv:2510.22344, Oct 2025

Abs arXiv Bib HTML PDF

FAIR-RAG is an evidence-driven agentic retrieval-augmented generation framework for complex multi-hop questions. Its Structured Evidence Assessment module audits accumulated evidence, identifies explicit information gaps, and guides targeted iterative retrieval before faithful answer generation.
@article{aghajaniasl2025fairrag, title = {FAIR-RAG: Faithful Adaptive Iterative Refinement for Retrieval-Augmented Generation}, author = {Aghajani Asl, Mohammad and Asgari-Bidhendi, Majid and Minaei-Bidgoli, Behrooz}, journal = {arXiv preprint arXiv:2510.22344}, year = {2025}, month = oct, archiveprefix = {arXiv}, primaryclass = {cs.CL}, }
arXiv
FARSIQA: Faithful and Advanced RAG System for Islamic Question Answering

Mohammad Aghajani Asl and Behrooz Minaei Bidgoli

arXiv preprint arXiv:2510.25621, Oct 2025

Abs arXiv Bib HTML PDF Website

FARSIQA is an end-to-end, safety-aware Persian question-answering system built on the FAIR-RAG architecture. It operates over 735,000 source records and 1.712 million indexed text chunks, combining scope validation, hybrid retrieval, iterative evidence refinement, and citation-grounded generation.
@article{aghajaniasl2025farsiqa, title = {FARSIQA: Faithful and Advanced RAG System for Islamic Question Answering}, author = {Aghajani Asl, Mohammad and Minaei Bidgoli, Behrooz}, journal = {arXiv preprint arXiv:2510.25621}, year = {2025}, month = oct, archiveprefix = {arXiv}, primaryclass = {cs.CL}, }