Tehran, Iran · Open to collaboration
AI software engineering · generative AI research
AI Engineer & Team Lead for Computer Vision and Generative AI at ModAI, building production LLM/VLM systems for multimodal retrieval and recommendation. Part-time at RamonAI building an agentic AI sales assistant. Previously a researcher at the University of Tehran on interpretable representation learning. Open to collaboration.
01 — Introduction
Research rigor, engineering discipline, product judgment.
I work at the intersection of generative AI research and scalable software systems. My background spans interpretable representation learning, multimodal retrieval, computer vision, recommendation, and trustworthy AI.
In industry I turn those foundations into services that are measurable, observable, and resilient — from embedding pipelines and hybrid search to routing, re-ranking, continuous evaluation, and model deployment.
I care about clean interfaces between models, data, and product logic; evaluation that reflects real user journeys; and systems that remain understandable as models and requirements evolve.
Offline and online metrics that mirror the actual user journey, not the convenient benchmark.
Clean seams between model, data, and product logic so the stack stays readable as it grows.
Research only counts once it survives autoscaling, peak load, and continuous release.
02 — Technical profile
From representation learning and multimodal modeling to search, infrastructure, and observability.
Production LLM/VLM services, CLIP-style alignment, diffusion models, VAEs, multimodal understanding, image editing, and virtual try-on.
Tool-using agents, workflow orchestration, routing, RAG, memory, tracing, and evaluation for multi-step user journeys.
Embedding pipelines, ANN/HNSW tuning, hybrid BM25 + vector retrieval, re-ranking, preference fine-tuning, and offline/online evaluation.
Containerized deployment, autoscaling, event-driven data planes, CI/CD, model versioning, continuous evaluation, and infrastructure automation.
SLIs/SLOs, tracing, metrics, logs, alerting, load testing, chaos testing, backpressure, and idempotent processing.
Interpretable and sparse latent spaces, disentanglement, contrastive objectives, robustness, explainability, and out-of-distribution behavior.
03 — Career
Leading applied AI from model to production.
Oct 2023
— Present
ModAI · AI-powered fashion discovery
Tehran, Iran
Lead CV/GenAI R&D for multimodal retrieval and recommendation across text, image, and hybrid experiences — owning services, data planes, deployment, and evaluation.
Feb 2025
— Present
RamonAI · Agentic AI sales assistant
Tehran, Iran
Build a robust, scalable agentic AI sales assistant that acts as a 24/7 admin for online shops — product discovery, Q&A, and sales workflows.
Sep 2023
— Feb 2025
University of Tehran · School of ECE
Tehran, Iran
Led TA teams across four graduate courses with Dr. M. Tavassolipour, Dr. M. A. Sadeghi and Dr. R. Hosseini — designing assignments, code frameworks and evaluation.
Feb 2022
— Sep 2024
Machine Learning & Computational Modeling Lab · University of Tehran
Tehran, Iran
MSc research on interpretable and disentangled latent representations in VAEs, diffusion models and GANs, including contrastive objectives such as angular and circle loss. Implemented all experiment tooling and reproducible training pipelines in PyTorch.
Dec 2022
— Sep 2023
NBIC · School of ECE, University of Tehran
Tehran, Iran
Built a CLIP-based emotion detection system aligning image and text representations, and collected and processed a 120-subject multimodal dataset.
Nov 2019
— Sep 2023
Freelance · Independent engagements
Tehran, Iran
Delivered freelance AI/ML projects spanning computer vision, deep learning, and data analysis; selected open-source work is available on GitHub.
04 — Academic work
My MSc research investigates structure in sparse latent spaces — connecting class-consistent active dimensions with global and class-specific factors.
Farshad Sangari Abiz · Reshad Hosseini · Babak N. Araabi
The work introduces a class-informed objective that encourages samples from the same class to share patterns of active latent dimensions, producing more structured and interpretable sparse representations.
University of Tehran · Advisers: Prof. Babak N. Araabi and Dr. Reshad Hosseini
05 — Portfolio
Practical work across trustworthy AI, generative modeling, and computer vision.
Investigating robust representations and classifiers under corrupted data using metric-learning objectives.
Implementations and experiments across variational autoencoder formulations.
Reconstructing images from intermediate CNN feature representations.
Predicting human visual attention using deep neural networks.
U-Net-based semantic segmentation for medical image analysis.
Hands-on teaching material for practical deep learning workflows.
06 — Background
Academic depth, national distinctions, and knowledge sharing.
University of Tehran · GPA 18.73/20 (3.89/4)
K.N. Toosi University of Technology · GPA 17.00/20 (3.64/4)
University of Tehran
Ranked 30th among more than 25,000 candidates
National academic distinction
Dr. M. Tavassolipour · Sep 2024
Dr. M. Tavassolipour · Jan 2024
Dr. M. Tavassolipour, Dr. M.A. Sadeghi · Sep 2023
Dr. R. Hosseini · Jan 2023
Dr. S. Haratizadeh · Sep 2023
Dr. M.-R. A. Dehaqani · Sep 2023
07 — Let’s connect
I’m interested in collaborations around generative AI, multimodal retrieval and recommendation, agentic systems, trustworthy AI, and scalable ML infrastructure.