Expertise in developing advanced NLP, speech and computer vision models, fine-tuning LLMs, and deploying scalable backend systems. Proven track record of delivering innovative solutions to complex challenges in data processing and AI applications.
Tehran | +989213207541 | [email protected] | LinkedIn: razi-taj-mazinani | GitHub: razi-tm
- PyTorch, TensorFlow, Keras, Fastai, LAVIS, YOLO, Scikit-Learn, OpenCV
- Familiar with:
- Dynamic programming, Memoization, Divide and conquer, Greedy, Sorting, Binary search
- Uninformed search strategies (Breadth-first search, Depth-first search, Uniform-cost search, Depth-limited search, Iterative deepening search)
- Informed search strategies (Greedy, A*)
- Local search (Hill climbing, Simulated annealing, Genetic algorithm)
- Adversarial search (Minimax, Alpha-beta pruning)
- Several indexing algorithms for similarity search
- Numpy, Pandas, Matplotlib
- Python, C/C++
- Assembling machine learning stations
- Linux, Git, Docker, Django, ElasticSearch, FAISS, PostgreSQL, Pgvector, Google Colab, Advanced VPN setup
- Full professional efficiency
- Machine Learning Engineer - Backend Developer at Sharif Search
- Machine Learning Engineer (Internship) at Sharif Search
- Problem-solving and critical thinking
- Communication and collaboration in cross-functional teams
- Time management and prioritization
- Creative thinking and innovation
- Adaptability and resilience in fast-paced environments
- Attention to detail and quality assurance
- Interpersonal skills for stakeholder engagement
- Curiosity and continuous learning
- A massive Persian dataset in question-answering format, including 50k samples and around 200k lines, tailored for NLP tasks.
- Fine-tuned Llama 3 using the dataset mentioned above. Enabled the model to understand and generate Persian text, overcoming its limitations for handling the Persian language.
- Developed a system that combines information retrieval and generative AI to improve the accuracy and relevance of generated text by leveraging external datasets.
- Implemented a robust pipeline for converting spoken language to written text with high accuracy, optimized for noisy environments.
- Designed and deployed a deep learning-based system to identify and distinguish speakers.
- Applied advanced neural networks to remove noise from audio signals, achieving superior clarity and fidelity.
Open-Source Contributor (Pyannote.Audio)
- Developed a fast and scalable image similarity search engine using embeddings.
- Innovated on traditional algorithms to accelerate face recognition tasks while maintaining high accuracy, optimized for large-scale datasets.
- Implemented data management strategies utilizing PostgreSQL to efficiently store, query, and analyze large datasets. Leveraged its advanced features, like indexing, to optimize performance and ensure scalability in data processing.
Deployment: Dockerized and deployed projects for production, ensuring scalability, reliability, and ease of integration when required.