AI Software Engineer · LLMs, AI Agents & Agentic Workflows
I am an experienced AI Engineer and Data Scientist with expertise in deep learning, NLP, and production-level AI applications. I design scalable ML models, AI-driven solutions, and LLM fine-tuning pipelines for real-world use.
I currently work as an AI Software Engineer at ADLAS in Vienna, Austria, building generative-AI document processing and conversational AI workflow builders. Previously, I held AI Engineer and Data Scientist roles at SnappShop, InoUp, and Metodata.
I am passionate about building AI-powered tools that drive real business impact — from AI agents and Retrieval-Augmented Generation to end-to-end AI system development.
Built an end-to-end AI-powered document processing solution using generative AI and OCR to extract structured and unstructured data from material specification files. Designed a conversational AI workflow builder that turns natural-language requests into fully automated workflows.
Developed an AI agent that autonomously enriches Elasticsearch and e-commerce product data with generative AI, improving search relevance and product discovery. Fine-tuned embedding models to improve similarity search in recommender systems via contrastive learning.
Designed a data-crawling infrastructure and service for the Jobinja platform and built an AI model to compute job-related metrics and answer employer inquiries about candidate resumes.
Built LLM-powered agents leveraging real-time Twitter insights for data-driven trading, and an Agentic RAG Twitter bot for automated content creation. Fine-tuned GPT-3.5, LLaMA2, and Mistral with RAG and LoRA; developed Persian stance detection and sentiment analysis models reaching 83% and 94% accuracy. Productionized models with FastAPI and Docker.
DigiNext
Designed an AI-driven social listening platform using NLP to analyze public sentiment and detect emerging trends, with deep-learning text analytics pipelines for large-scale data.
Conducted research on deep learning for NLP and developed a novel AI-based legal assistant system that improved accuracy and efficiency in legal document analysis.
Università degli Studi di Padova
Developed a novel matrix-completion approach using Radial Basis Functions (RBF) and an optimal Truncated SVD (TSVD) to enhance recommender systems.
Programming Languages
Frameworks & Tools
Focus Areas
Kharazmi University
GPA: 3.6 / 4.0
M. Alizadeh, M. Kubli, Z. Samei, S. Dehghani, M. M. Zahedivafa, J. D. Bermeo, M. Korobeynikova, F. Gilardi
Journal of Computational Social Science, 2025
M. Alizadeh, Y. Asgari, Z. Samei, S. Yari, S. Dehghani, M. Kubli, D. Zare, J. D. Bermeo, V. Batzdorfer, F. Gilardi
IEEE/ACM Int. Conf. on Advances in Social Networks Analysis and Mining (ASONAM), 2024
S. Dehghani, M. M. Zahedivafa, Z. Baghshahi, D. Zare, S. Yari, Z. Samei, M. H. Aliahmadi, M. Abbasi, S. Mirzamojtahedi, S. Ebrahimi, M. Alizadeh
Multidisciplinary Int. Symposium on Disinformation in Open Online Media (MISDOOM), 2024
See the full list on Google Scholar.
Why even the most capable models stumble without the right scaffolding — and how the agent harness makes them reliable.
Hard-won lessons on what it actually takes to make AI agents reliable, useful, and production-ready.
A clear look at the hardware behind trillion-parameter models and the systems that train them.
Why the future of AI belongs to people who understand systems, not just scale.
Read more on Medium.
I'm always open to discussing research, collaborations, or interesting ideas in AI. The best way to reach me is by email, or connect with me on LinkedIn.