000 Computing Genealogy Project Meta-science Engineering Bachelor's capstone
This bachelor's capstone project develops the Computing Genealogy Project into a usable public-facing system for exploring academic relationships in computing. The work includes collecting and cleaning advisor-student and collaboration data, improving the UI/UX for search, browsing, and visualization, designing stable APIs for programmatic access, and supporting embeddable views that can be integrated into personal pages, lab websites, or course materials.
100 Hypervisor Security SoK Full lifecycle system security Engineering Master's capstone Hiring for 2026-2027
This master's capstone project studies hypervisors from three connected angles: hypervisors used as security mechanisms, the security of hypervisors themselves, and the risks introduced by insecure hypervisor designs or deployments. The project aims to build a systematic knowledge base that clarifies threat models, isolation assumptions, vulnerability classes, measurement methods, and open research problems across commodity, cloud, embedded, and security-oriented virtualization systems.
101 Firmware Rehosting Service and Reports Full lifecycle system security Engineering Master's capstone
This master's engineering project builds a firmware rehosting service that turns firmware images into reproducible emulation and analysis tasks. The system should support firmware intake, unpacking, architecture and platform identification, environment reconstruction, service launch, instrumentation, and execution monitoring. It also produces structured reports that summarize rehosting status, reachable services, runtime behavior, configuration assumptions, observed failures, and follow-up security testing opportunities.
102 Hypervisor Fuzzing Service and Reports Full lifecycle system security Engineering Master's capstone
This master's engineering project builds a hypervisor fuzzing service for repeatable testing of hypervisor interfaces and virtualization components. The system should manage target setup, guest workload generation, interface modeling, input mutation, crash detection, coverage collection, triage, and regression replay. It also produces structured reports that describe tested interfaces, fuzzing campaigns, discovered crashes, reproducibility, root-cause hypotheses, security impact, and recommended next steps.
103 Building a Recursive Self-Improvement Workflow for POC Generation Cybersecurity LLMs, AI agents, and benchmarks Engineering Master's capstone Hiring for 2026-2027
This master's engineering project builds a recursive self-improvement workflow for proof-of-concept generation. Based on Magma, ground-truth bug benchmarks, and a workflow DSL, the project designs and evaluates an agentic pipeline that can generate POCs, execute them against known vulnerable targets, diagnose failures, revise its workflow, and accumulate reusable strategies over time. The goal is to turn benchmark-backed POC generation into a measurable engineering loop for studying how execution feedback can continuously improve security tools.
104 Paper2Product: A Research-to-Technology Transfer Workflow Meta-science Engineering Master's capstone
This master's engineering project studies how the central challenge of academic work changes when publishing papers is no longer the main bottleneck: the problem shifts from producing visible research to turning research into knowledge that can be verified, reused, and deployed. The project develops Paper2Product, a workflow that ingests a large research corpus, extracts claims, evidence, methods, data, and artifacts, groups related findings, identifies practical needs, and proposes feasible paths toward tools, datasets, services, standards, or follow-up studies. It also investigates why technology transfer is difficult, including missing context, weak reproducibility, tacit knowledge, intellectual-property constraints, incentive misalignment, and the gap between research prototypes and reliable systems. The deliverable is a working prototype, a documented transfer workflow, and case studies that evaluate where automation helps and where human judgment remains necessary.
105 Agent for Paper Reproduction and Artifact Improvement Meta-science Engineering Master's capstone
This master's engineering project develops an agent for automating artifact evaluation, paper reproduction, and artifact improvement. Given a paper and its available materials, the agent identifies the claimed experiments and evaluation criteria, generates missing artifacts when feasible, configures and runs the experimental environment, compares reproduced results with the paper's reported results, diagnoses discrepancies and failures, and iteratively improves implementations, configurations, and evaluation metrics. The goal is to make artifact evaluation more reproducible, scalable, and useful while clearly recording the evidence, assumptions, and human interventions behind each result.
106 SoK Generation Workflow Meta-science Engineering Master's capstone
This master's engineering project builds a workflow for producing high-quality Systematization of Knowledge (SoK) papers and surveys. The workflow supports literature discovery, corpus construction, relevance screening, taxonomy development, evidence extraction, claim synthesis, gap analysis, and iterative drafting. It integrates with Project 105 to reproduce experiments and evaluate research artifacts, keeping the survey's narrative, structured data, reproduced results, tables, figures, and conclusions consistent with one another. The goal is to make SoK production more systematic and auditable while preserving the expert judgment required to define scope, resolve conflicting evidence, identify meaningful research gaps, and construct a coherent intellectual contribution.
107 Paper Diff and Duplication Detection Workflow Meta-science Engineering Master's capstone
This master's engineering project builds a large-scale workflow for comparing papers and detecting duplicated or insufficiently differentiated research. Inspired by binary diffing, the workflow develops a semantic paper-diff representation that aligns research questions, assumptions, methods, datasets, experimental settings, results, claims, and artifacts across papers. It identifies content that is equivalent or substantially similar, evidence and conclusions that contradict one another, and genuinely new material introduced by a later work. The system combines scalable retrieval, candidate-pair generation, structured extraction, semantic alignment, contradiction analysis, provenance tracking, and human-review interfaces to analyze large research corpora. The goal is not merely to detect textual overlap, but to distinguish reuse, replication, incremental extension, conflicting evidence, and substantive novelty with traceable supporting evidence.
200 Regression-Aware LLMs for System Programming Cybersecurity LLMs, AI agents, and benchmarks Research PhD project
This PhD research project investigates how large language models are changing system programming and how they introduce regressions into evolving codebases. The project tracks emerging LLM-based system-programming workflows, develops methods to detect functional, performance, security, and reliability regressions, and studies how regression risk accumulates during long-running software development and maintenance. It further explores regression-aware reflection and self-evolution mechanisms that allow LLM-based agents to learn from failures, preserve previously validated behavior, and avoid reintroducing regressions over time.
201 Fault Injection and Prevention Full lifecycle system security Research PhD project
This PhD research project studies fault injection as a practical attack path when software systems become increasingly difficult to exploit directly. The project examines how attackers use voltage, clock, electromagnetic, laser, and other physical fault models to bypass security checks, corrupt control flow, extract secrets, or undermine trusted execution. It further develops prevention and detection techniques across hardware, firmware, compilers, runtimes, and system software, with the goal of building systems that remain secure even under realistic fault attacks.
202 Cyberspace Range Cybersecurity LLMs, AI agents, and benchmarks Research PhD project
This PhD research project builds a cyberspace range that mimics realistic cybersecurity environments at scale. The range combines complex infrastructure, strategic multi-step missions, realistic attacker and defender workflows, and large numbers of predefined agents to create reproducible yet dynamic cyber scenarios. The goal is to exercise and evaluate cyber models in settings closer to real operations, where success depends not only on isolated vulnerability solving, but also on planning, adaptation, coordination, situational awareness, and long-horizon decision making.
203 Defense in Depth for Sandboxes Full lifecycle system security Research PhD project Hiring for 2026-2027
This PhD research project studies defense-in-depth architectures for process-based and VM-based sandboxes. The project investigates how sandbox escapes and policy bypasses emerge across system-call interfaces, shared resources, runtime services, and hardware boundaries, then develops layered defenses that combine behavioral monitoring at security-relevant interfaces, exploit prevention, and compartmentalization. It further explores techniques such as pattern matching, hardware-assisted enforcement, and hardware-software co-design to make sandboxed execution more resilient against both known attack patterns and previously unseen exploitation strategies.
204 Aero-Space Security Across the Stack Full lifecycle system security Research PhD project
Today’s aero-space systems connect satellites, aircraft, drones, base stations, and cloud control planes without a shared trust floor. A fault or compromise at one layer can cascade across the stack, while GPS spoofing, jamming, avionics intrusion, and weak cross-domain authentication threaten safety and mission continuity. This project builds a layered security survey and reference architecture: map representative targets across altitudes, organize attacks and mitigations at hardware, protocol, and system layers, and make cascading paths from cloud and base stations to aircraft, satellites, and end devices explicit. The study connects trusted execution, integrity verification, and secure orchestration, and examines whether defenses compose end to end.
300 AI-Hard Problems in System Security Full lifecycle system security Open problem Bachelor's project Master's project PhD project
This open problem asks whether system security contains fundamental challenges—and compositions of those challenges—that reliably cause current AI models and agents to fail. Software hardening and resistance to reverse engineering provide a representative setting: obfuscation, packing, anti-debugging, anti-emulation, self-modifying code, opaque predicates, indirect control flow, and adversarially composed transformations can each obstruct analysis, while their combinations may create qualitatively harder problems than any component alone. The project builds a taxonomy and benchmark suite of primitive and composed AI-hard tasks, measures where models lose correctness, robustness, efficiency, or the ability to recover from mistakes, and studies whether failures arise from missing knowledge, limited context, long-horizon reasoning, tool use, partial observability, or adversarial structure. Undergraduate projects can construct and evaluate focused challenge instances, master's projects can build scalable benchmark and analysis systems, and PhD projects can characterize capability boundaries, compositional hardness, and new methods that overcome them.
