Curious enough to research it. Practical enough to ship it.

• Building the next generation of personal assistants for businesses and professionals using AGI and agentic AI.

RadarCX ↗

• Contributed to the rapid design, development, testing, and integration of a proprietary Text-to-SQL AI system while applying recent research approaches.
• Revived a context-classification model that had plateaued for over a year by applying advanced regularization, a custom architecture, and fine-tuning techniques.
• Increased test-set accuracy by 13.7%, breaking through the model’s long-standing performance ceiling.
• Improved runtime efficiency by 13.4× without increasing model size or infrastructure overhead.
• Reduced calibration error by 3.99% on a log scale, improving reliability and reducing overconfident misclassifications.
• Implemented class weighting and hard-negative mining to address dataset imbalance and improve learning from difficult cases.
• Developed training-diagnostic visualizations to improve experiment tracking.
• Integrated Expected Calibration Error (ECE) to assess and improve prediction confidence.
• Reduced training VM costs by 92.32% through performance optimization and operating-system-level tuning.
• Refactored the code for modularity and extensibility, exposing key hyperparameters through a command-line interface for easier experimentation.
• Applied machine-learning design patterns and coding practices, producing clean documentation to accelerate onboarding and maintenance.
• Integrated the improved ContextClassifier into the DIY configuration phase for a smoother user experience.

• Designed and implemented normalized database tables using DBMS practices, reducing data footprint and improving consistency.
• Developed and maintained a MySQL database with SQLAlchemy for efficient storage and retrieval between the frontend and backend.
• Built the frontend with Ninja, HTML, CSS, and JavaScript, improving responsiveness and the user experience.
• Engineered Flask backend routes and improved API data handling.

• Developed a language model using Graph Attention Networks for text summarization.
• Cached tokenized inputs to handle large datasets while reducing RAM use and runtime.
• Vectorized operations to accelerate matrix multiplication.
• Used penalty-based methods to handle class imbalance.
• Improved the selection F1 score from 35% to 38% through multiple machine-learning approaches.
• Parallelized training and fine-tuning, reducing runtime by 75%.
• Led an internship team that generated a dataset in two months after an earlier four-month attempt had been unsuccessful; distributed processing reduced one stage from 29 days to four days.
• Developed a model that outperformed models reported by the Microsoft Dynamics 365 AI Research team and Facebook.

Published paper ↗

• Assisted the principal lab instructor in Practical Software Development.
• Taught pointers and assisted labs on software design patterns, Doxygen, and GoogleTest; other topics included anti-patterns, Git, refactoring, and UML.
• Selected for the role as one of 30 candidates based on previous performance.

• Instructed Fundamentals of Programming II labs.
• Designed lab lectures and exercises and independently recorded extra tutorial videos for students.
• Promoted to the role following strong student evaluations of teaching and communication in the previous position.

Lectures and exercises ↗Tutorial videos ↗

• Instructed Fundamentals of Programming I labs.
• Used C++ to teach 50 students in a multicultural environment and voluntarily mentored students in the Computer Science program.
• Independently developed a Python program to automate grading, saving approximately 100 hours.

• Selected from a competitive applicant pool to research and develop chatbot systems.
• Participated in weekly research-group meetings and collaborated with other researchers.
• Used GPT-1 and CLIP to develop a multi-domain, multimodal chatbot system that improved image search by 60%.
• Used a Singapore tourism dataset to design a travel-assistant chatbot.

• Learned computer-network concepts, backup tools, and internal operational systems.

• NSERC funding — recipient: research supervisor.
• University of Lethbridge Graduate Research Award — recipient: Reza Ardestani; nearly CAD 20,000.
• Alberta Innovates award — recipient: Reza Ardestani; nearly CAD 30,000.

Read the publication ↗

• Implemented projects on code completion, aspect-oriented programming, and REST APIs.
• Used an agile development pipeline, static analysis, GoogleTest, and code-coverage tooling.
• Worked with DevOps practices, Docker, and Kubernetes.

Project repository ↗

• The team learned an unfamiliar development stack quickly and finished as the second-best team in a competitive class.
• Built the website using Scrum, continuous integration and delivery, and push-notification technology.

Project repository ↗

• Covered heuristic search, genetic algorithms, adversarial search, knowledge-based systems, and evolutionary algorithms.

Course projects ↗

• Used AR, ARIMA, and SARIMAX from statsmodels and an LSTM model from TensorFlow.

Course repository ↗

• Used the Amazon movie dataset and Latent Dirichlet Allocation (LDA) to extract topics.
• Detected sentiment using both unsupervised and supervised approaches.
• Other projects included Twitter text classification, Persian-language spam detection, and web crawling.

Final project ↗Other course projects ↗

• Completed 12 Linear Optimization problem sets implemented in LINGO.
• Implemented six industrial optimization problems in GAMS for Combinatorial Optimization and Network Analysis.

Linear Optimization ↗Combinatorial Optimization ↗

• Database Management Systems: A+
• Data Mining: A
• Advanced Natural Language Processing: A
• Advanced Software Engineering: A−

• Artificial Intelligence: A+
• Data Mining: A+
• Social Media Computing: A+
• Data Structures and Algorithms: A+
• Logic and Set Theory: A+
• Software Design: A+
• Theory of Computation: A+
• Computer Geometrical Design: A

Earned full marks in Calculus and Differential Equations on nationally administered university-admissions examinations for high-school students.

Received the Upper Bound Talent Bursary to attend the Upper Bound conference in Edmonton, a global hub for reinforcement learning and home to Turing Award-winning reinforcement-learning scientists.

Received a highly competitive award recognizing innovative scholars advancing AI and innovation across multiple domains.

Received the award twice for outstanding research and academic achievement at the University of Lethbridge.

Received a Coca-Cola-sponsored master’s entrance scholarship recognizing academic excellence and potential for future leadership.

Ranked in the 79th percentile in Quantitative Reasoning among GRE test takers worldwide.

Ranked in the top 1% of Iran’s national university entrance examination and entered Amirkabir University of Technology.

Passed NODET’s annual nationwide selection examination for students entering secondary school; its schools provide more advanced education across subject areas.

Led a well-received Google Developer Groups session on data handling, model training, optimization, and machine-learning project structure, with high attendee satisfaction. Also prepared supplementary material and YouTube tutorials for graduate students and machine-learning researchers covering the full ML project pipeline.

GDG event ↗Supplementary material ↗

Helped rebuild the club through weekly constitution-writing meetings; organized Nowruz and Yalda festivals, an orientation for incoming Iranian students, and events raising awareness of women’s rights in the Middle East.

Maintained computer-lab systems and required course software for two semesters; helped students use on-site printers and connect to the network.

Served as liaison between students and university administration, expedited facility requests, addressed student concerns, and helped organize campus events including the Ramadan Festival.