sep 2026 — present · ongoing
Deciding Without Generating: Jev at the Decision Points of LLM Systems
non-generative deciders vs llms where an llm system decides rather than writes
- llm agents
- τ²-bench
- benchmarking
- routing
- semantic caching
- energy
<research>
//software analysis · llm & agentic systems · empirical evaluation
//interests
I am interested in how we know whether software systems actually work and what they actually cost: learned models for software analysis (bug finding, code classification, test generation), the efficiency of LLM and agentic systems, and the empirical side that makes both assessable: ground truth, benchmarks, and honest evaluation of what these tools do in real deployments.
//work
2026 · manuscript in preparation
Silent Ground: Adversarial Collection Suppression Against Attack Path Management
how far an adversary can quietly degrade the data bloodhound reasons over
- active directory
- bloodhound
- adversarial data
- attack graphs
- go
fall 2024 — spring 2025 · presented at the Purdue Undergraduate Research Expo, fall 2024
Fuzzing in Open-Source Software: Current State, Evolution, and Challenges
an empirical study of fuzzing practice across the top 10,000 oss projects
- fuzzing
- oss-fuzz
- fuzz introspector
- empirical se
- ground truth