Hassan El-Sheikha

Upper-year coursework:

Operating Systems (CSC369), Compilers and Interpreters (CSC488), Information Security (CSC347), Computer Security (CSC427), Machine Learning (CSC311), Computer Networks (CSC358), Algorithms (CSC373), Computability Theory (CSC363), Functional Programming (CSC324), Databases (CSC343), Web Programming (CSC309), Artificial Intelligence (CSC384).

Research courses:

CSC492 on advanced computability and Turing machines; CSC392 and CSC493 on the development of a Turing machine description language; CSC393 on the development of an LMM-powered grading assistant.

• Developed LLVM-based analysis and transformation passes for SmartHLS, a C++-to-Verilog high-level synthesis compiler.
• Designed and implemented a system for debugging arbitrary-precision number types, reducing customer development time while lowering area and power usage in real HLS designs. This work led to the U.S. patent application Systems and Methods for Control of Arbitrary-Precision Numbers in High-Level Synthesis.
• Built a retrieval-augmented generation framework for semantic search and question answering over multimodal data, including code, documentation, and training materials, which was adopted by multiple business units across the company.
• Contributed to a method for optimizing HLS-generated Verilog using large language models, leading to the U.S. patent application Systems and Methods for Optimizing HLS-Generated HDL Code using ReAct Agents with Techniques for Overcoming Generative AI Model Output Context Window Limitations.
• Collaborated with compiler, verification, and product teams; authored technical design documents and presented prototypes to engineering leadership and patent counsel.

• Designed and implemented SiteWatch, a monitoring and alerting tool for Islandora/Drupal-based digital preservation systems, enabling librarians to track ingest jobs and repository health.
• Extended and maintained open-source tools, including Islandora Workbench, to support large-scale metadata ingest, validation, and transformation for historical newspaper collections.
• Worked with librarians and historians to design pipelines for extracting, structuring, and preserving archival content for digital humanities research.

• Designed and implemented a Manifest V3 browser extension and Python FastAPI backend to integrate Large Multimodal Models into university grading platforms, automating detailed feedback for handwritten student submissions while keeping a human in the loop.
• Evaluated the system on 405 submissions, achieving a 0.95 Concordance Correlation Coefficient with expert graders and reducing batch grading time by 96%, from 75 minutes to 3 minutes, through parallel API orchestration.
• Authored a manuscript, Towards Scalable Retrieval Practice: Integrating LMMs into Established Grading Platforms via a Browser Extension, proposing a privacy-preserving framework for high-frequency assessment in large computer science courses.

• Designed and implemented Varphi, a domain-specific language for specifying and simulating Turing machines and finite automata, with support for multi-tape computation and modern developer tooling.
• Built the language frontend with ANTLR and developed a modular Python-based compiler framework with multiple backends, including a Debug Adapter Protocol–compliant runtime.
• Developed a Visual Studio Code extension for Varphi with syntax highlighting, interactive debugging, and tape visualization.
• Helped deploy Varphi in University of Toronto computability courses in both Winter 2025 and Winter 2026, where it was used by roughly 180 students in each offering.
• This work led to a peer-reviewed WCCCE 2025 paper and a follow-up WCCCE 2026 paper on modern tooling for teaching Turing machines.

Links: GitHub, Documentation, WCCCE 2025 Paper, News