I'm a Computer Engineering graduate from the University of Tehran, interested in graduate study and research in artificial intelligence. My projects span machine learning, computer vision, natural language processing, and software systems.
I enjoy turning ideas into working implementations and evaluating how well they perform, especially when learning models need to produce reliable results.
- Machine learning: deep learning, reinforcement learning, and reliable AI.
- Language and vision: natural language processing, computer vision, and multimodal learning.
- AI systems: information retrieval, retrieval-augmented generation, and language model evaluation.
Best Undergraduate Project Award, Project Day, Faculty of Electrical and Computer Engineering, University of Tehran, 2026.
Supervised by Prof. Yadollah Yaghoobzadeh.
A corpus-grounded assistant that uses Gemma to interpret requests and BGE-M3 to retrieve Persian expressions. The final proverb wording comes from a reviewed corpus.
- 1,820 expression families and 2,067 surface variants.
- 81% end-to-end accuracy on a benchmark of 200 prompts.
- In a blind comparison of 185 prompts, responses containing fabricated or unattested expressions fell from 31.35% with base Gemma to 1.08% with the proposed system.
The repository includes the implementation, corpus, evaluation results, and a Persian RTL interface.
| Project | Focus and results |
|---|---|
| Persian Legal QA with RAG | Question answering over 120 scanned pages using OCR, multilingual embeddings, and retrieval. RAGAS answer relevancy: 0.78. |
| Tool-Using Travel Assistant | 7 tools with structured tool calling and fallback handling; 14 test cases for tool selection and argument routing. |
| Instruction Tuning via LoRA | Fine-tuned TinyLlama using approximately 0.2% of parameters. IFEval instruction-level accuracy improved from 0.09 to 0.26. |
| U-Net Image and Video Inpainting | Image inpainting on CelebA with PSNR 27.52 dB and SSIM 0.923, extended to video frames. |
| Spoken Language Identification | Audio features and classical classifiers evaluated on 712 clips in 4 languages. |
I also build software and embedded systems: Wi-Fi Indoor Positioning and WealthWise.
Programming: Python, C/C++, Java
Machine learning: PyTorch, scikit-learn, NumPy, Pandas
Language models and retrieval: Hugging Face Transformers, PEFT/LoRA, LangChain, LangGraph, LanceDB, RAGAS
Software and systems: Spring Boot, REST APIs, GitHub Actions, Docker, Linux
- B.Sc. in Computer Engineering, University of Tehran, 2026.
- Teaching Assistant, Machines and Language Theory, University of Tehran, 2024.
- Rank 305 among approximately 160,000 participants in Iran's national university entrance examination, Mathematical Sciences, 2020.
Open to graduate study and research opportunities across AI and related fields.