AI/ML Engineer building recommenders, agents, and pipelines.
CS Student & AI/ML Engineer, Tehran
I build AI systems that turn research into practical tools.
My work spans machine learning, large language models, recommender systems, and intelligent automation, with a focus on creating software that solves real problems rather than showcasing algorithms.
My journey combines research, engineering, and education. I've contributed to AI research, mentored hundreds of students as a TA, reviewed academic papers for the department journal, and continuously explore new developments in intelligent systems.
I believe great software comes from balancing curiosity with engineering discipline. I like exploring new ideas, but I care just as much about clean implementation, maintainability, and building products that deliver real value.
Selected work
What I do.
Building intelligent systems with PyTorch, LangGraph, LangChain, N8N, Scikit-learn, and LLM APIs. Specializing in AI Agents, recommendation systems, NLP, and predictive models.
End-to-end data pipelines using pandas, NumPy, PowerBI and SQL. Experience with large record datasets and statistical analysis.
Landscape and travel photography. Published on Unsplash with 2M+ views, capturing the world through my lens.
Scientific reviewer for CS Department Journal. Published section on ML in Mental Health and Suicide Risk Detection.
Full-stack deployment with Docker, Git, Streamlit, and cloud APIs. Built Telegram bots and web applications.
Head TA across 27+ course credit hours. Supervisor & Teacher at Salam High School, Head of Scientific Committee, coordinating seminars and academic events.
Engineering, teaching, mentoring
- Built a hybrid RAG engine (vector + keyword search + re-ranking) for AI medical triage and specialist routing, grounded in a large clinical knowledge base of textbooks and cases.
- Fine-tuned LLMs and rebuilt the knowledge base, cutting retrieval latency by about 70% and raising triage accuracy to roughly 90%; served multiple models through the NVIDIA API for low-latency inference.
- Developed the website and automation scripts with three portals (patient, insurer analytics, admin) for an insurance-sector investor briefing.
- Led and taught across 27 credit hours of undergraduate CS courses as TA & Head TA, coordinating labs, grading, and office hours.
- Supervised and taught programming and data science courses at Salam High School, mentoring hundreds of students.
- Head of the Scientific Committee, organizing seminars, academic events, and departmental initiatives.
- Scientific reviewer for the CS Department Journal, contributing a section on ML in Mental Health & Suicide Risk Detection.
Notes & post-mortems
Bubble sort is famously useless — except in three niche cases: tiny arrays, real-time rendering, and sorting animations.
Apple's new paper shows reasoning models collapse on hard logic puzzles — and the debate over why is still raging.
MIT's new report says 95% of generative AI pilots fail to deliver measurable value. Here's why — and what the 5% do differently.