Projects پروژه‌ها

Research demos and systems, plus earlier course and robotics work.

The largest open Persian speech corpus (3,500+ hours, 1,800+ speakers) with an interactive zero-shot voice-cloning demo. Corpus, pipeline, and fine-tuned models.

8,000+ Persian audio samples across 16 tasks — ASR, translation, QA, emotion, and culturally grounded tasks like poetry meter (vazn) and Dastgah music; 10 tasks are the first of their kind in any language.

Evaluating LLMs on Persian-English medical question answering. LREC 2026.

A 17M-sample dataset and BERT-based models for Persian punctuation restoration.

Multi-platform chatbot (Telegram, WhatsApp, web) with a custom RAG pipeline; 20,000+ active users on the University of Tehran website.

Open-source architecture for a low-cost assistant robot: real-time facial emotion recognition, gesture responses, speech commands. ICRoM 2022.

Multimodal QA for animal counting 2025

Synthetic multi-object counting dataset of Iranian animal species (Google Search API + SAM2) and a multimodal QA model, designed as an ML course contest at Sharif.

Earlier projects