Open to opportunities

Mohammad Hossein Abbaspour

@mohammadhosseinabbas

Message

I build retrieval-augmented and multi-agent LLM systems for reliable question answering.

Iran

Message

I'm building hybrid retrieval-augmented legal QA and multi-agent assistants at Faranegar Management & Technology Consulting, improving citation accuracy and answer traceability through semantic and keyword retrieval with reranking. I also evaluate GPU memory and throughput requirements to support hardware-selection decisions.

Previously, I built an end-to-end RAG QA system at Dadmatech and automated LLM content-analysis workflows at NC State University. My research at IUST spans multi-agent LLM policy compliance, parameter-efficient fine-tuning, and an MPNet system for SemEval-2024 Task 9 that improved on the baseline by 11%.

Experience

Work history, roles, and key accomplishments

Current

AI Specialist

Current

Faranegar Management & Technology Consulting

Sep 2025 - Present (11 months)

Building a hybrid retrieval-augmented legal QA assistant combining semantic and keyword retrieval with reranking for improved citation accuracy. Developing a multi-agent assistant for central-bank employees and evaluating GPU infrastructure requirements.

RAG Reranking Multi Agent Systems GPU Infrastructure Memory Estimation

AI Software Engineer Intern

Dadmatech

Jul 2025 - Sep 2025 (2 months)

Built an end-to-end RAG-based QA system with OpenAI Vector Stores, covering ingestion, indexing, retrieval, and response generation. Designed a teacher-student QA architecture to support iterative answer generation.

RAG Ingestion Indexing Teacher Student Architecture

Undergraduate Research Assistant

Iran University of Science and Technology

Oct 2023 - Jul 2025 (1 year 9 months)

Investigated policy compliance, deceptive interactions, and countermeasures in multi-agent LLM systems for a B.Sc. thesis. Explored parameter-efficient fine-tuning methods for mixture-of-experts models and co-developed an MPNet system for SemEval-2024 Task 9, achieving an 11% improvement over baseline.

Policy Compliance Deceptive Mixture Of Experts SemEval

Research Assistant Intern

NC State University

Dec 2024 - Apr 2025 (4 months)

Automated LLM-based content-analysis workflows and designed statistical evaluations for a multi-agent research system. Refined prompts and model configurations to improve reliability and quality of LLM outputs.

LLM content analysis Multi Agent Systems Prompt Engineering Configuration