Manager: Fizza Ahmad Sheikh

• Executing Product Data Science initiatives within the Support Data Team.

Manager: Taryn Govey and Fizza Ahmad Sheikh

• Built LLM-as-a-judge at scale of millions of interactions for customer service conversations.

• Designed and deployed airflow dags for support tickets evaluation.

• Owned Search and Retrieval product on help.shopify.com, delivering insight to improve the system performance, and drive experimentation on product.

Field:  Generative AI and Agent Development
Project Manager: Nileena Thomas

• Benchmarked PDF parsers including Docling, MinerU, and VisionParser on both content and layout accuracy

• Benchmark LLMs & LRMs through OpenRouter including Gemini, Claude, and GPT on HTML code generation and layout understanding

• Designed and Developed 3 agents including a sequential workflow, a single-agent with multiple tool, and a multi-agent system using Google ADK for converting PDF to Django-compatible and dynamic HTML file

Field:  Natural Language Processing & Information Retrieval
Manager: Prof. Ebrahim Bagheri

• Implemented text processing pipelines and web crawler to extract and transform data

• Designed and developed Peerispect, a claim verification tool for research peer review using open-source LLMs such as qwen3, deepseek-r1, and llama3 via the Ollama framework, and deployed the service using FastAPI [demo]

• Designed and developed PeeriScope, a review quality dashboard to analyze reviews and visualize metrics computed using LLM inference and NLP methods via the HuggingFace library [demo]

Field: Computer Vision Security
Advisor: Prof. Mohamed Hefeeda

• Researched multimedia forensics and computer vision security focusing on images

• Developed deep-based models for image watermarking, provenance, and authentication

• Designed a robust and scalable system to tackle challenges in online image manipulation detection like social media platforms to fight against misinformation

Field: Applied Machine Learning
Advisor: Prof. Abbas Heydarnoori

• Transformed KGTorrent dataset with 250,000 Jupyter Notebooks to structured dataframes

• Extracted pre-defined features from dataframes to investigate code comprehension

• Evaluated feature effectiveness in code understandability and created importance map

Field: Applied Deep Learning
Adviser: Prof. Abbas Hosseini

• Cleaned and analyzed a large dataset with 10 millions of rows and 300 feature columns

• Expanded the audience base, via user feature similarity and collaborative filtering

• Trained and tested a deep model to boost revenue and optimize ads via personalized approach