Computer engineer (Michigan State, B.S. CompE, Dec 2025) building robotics simulation, test and autonomy software and evaluated ML systems: LLM fine-tuning and serving, grounded RAG, agent prompt-injection benchmarks and GPU-parallel RL.
Currently: Associate (Java Full Stack Developer), Revature, Jun 2026–present
Links: Portfolio | Resumes | LinkedIn | Email | Handshake | LeetCode
Built with an AI coding agent under spec-driven development; every number below comes from a seeded script in the repository.
| Project | What it is | Headline result |
|---|---|---|
| Autonomy Test Harness ROS 2 Jazzy, Nav2, Gazebo Harmonic, Docker, Jenkins |
Scenario-based test system for Nav2 on TurtleBot 4: seeded scenario generator, lidar fault injector, ground-truth-only oracle | 210 headless runs (30 scenarios × 7 conditions): Nav2 passed 100% unfaulted, 80% under sector blackout, 0% under scan dropout or freeze |
| Sensor Sim Sandbox Unreal Engine 5.8 (C++), ROS 2 Jazzy, rosbridge, Python |
Drivable UE 5.8 rover whose lidar, camera, IMU and GPS share a seeded noise/dropout/latency pipeline and stream to ROS 2 | Measured lidar std 0.0200 m vs 0.02 m configured; 33 Unreal automation tests |
| Learned Opponents PyTorch, MuJoCo Warp, rsl-rl PPO, ONNX, Unreal Engine 5.8 |
PPO navigation policies trained in 4,096 GPU-parallel arenas, exported to ONNX and run as the opponent in a UE 5.8 game | With 4 control steps (200 ms) of latency on both action and observation: 0.131 success (baseline) vs 0.896 (domain-randomized) |
| Project | What it is | Headline result |
|---|---|---|
| Grounded QA Service FastAPI, Granite on Ollama, cross-encoder reranking, Docker |
RAG service that answers only from retrieved passages, validates every citation and refuses below a tuned confidence threshold | Recall@5 0.847 → 0.957 with hybrid retrieval + reranking (150 answerable SQuAD v2 questions) |
| Fine-Tune and Serve PyTorch, PEFT (LoRA), llama.cpp, Docker |
LoRA fine-tune of Qwen2.5-1.5B-Instruct for intent-and-slot extraction, quantized to GGUF and served behind an OpenAI-compatible endpoint | Field-level F1 0.850 ± 0.002 (3 seeds) vs 0.358 few-shot base; Q4_K_M keeps F1 0.846; peak whole-GPU memory 4859 → 2617 MiB |
| Agent Injection Bench AgentDojo, llama.cpp, OpenAI API, DeBERTa |
Reproducible benchmark of how often tool-using LLM agents follow instructions planted in tool outputs, and what each defense costs | 7,200 cases (hosted model: 2 of 5 styles): a DeBERTa injection filter cost 26.5 to 56.2 points of task success for at most a 2.3-point drop in attack success |
ECE-480-Team-12-TI-EMG-AWS · C, TI Tiva C (TM4C123) · demo video
Sole software member of a four-person Texas Instruments-sponsored senior design team (Fall 2025): wrote the TM4C123 C firmware: ADS131M02 SPI ADC driver, SSD1351 OLED graphics, menus and four game modes, and a baseline-corrected EMG measure.
- rag-lite-tfidf-eval: retrieval library (TF-IDF/BM25/embeddings, optional reranking) with IR metrics; Grounded QA Service builds on it
- robotics-2d-perception-control: 2D differential-drive sim with an OpenCV perception pipeline, closed-loop controller and state machine
- ai-sim-rl-cartpole: PPO on CartPole with multi-seed evaluation and CI
- cse482-reddit-events: Reddit event detection and extractive summarization with ROUGE evaluation
- roblox-ai-game-repo: Lua NPC chat backed by an external LLM, with text filtering and rate limiting
- Languages: Python, C++, C, Java, SQL, TypeScript, MATLAB, Lua
- Robotics & simulation: ROS 2 Jazzy, Nav2, Gazebo Harmonic, rosbridge, Unreal Engine 5.8, MuJoCo Warp, OpenCV
- ML & LLMs: PyTorch, Hugging Face Transformers, PEFT/LoRA, llama.cpp, GGUF, Ollama, ONNX Runtime, PPO (rsl-rl, Stable-Baselines3), domain randomization
- Evaluation: hybrid retrieval, cross-encoder reranking, Recall@K, nDCG, F1, bootstrap CIs, LLM-as-judge
- Embedded: TI Tiva C (TM4C123), Code Composer Studio, sEMG acquisition and signal features
- Testing & CI/CD: pytest, Hypothesis, Unreal Automation, JUnit 5, Cucumber, Jenkins, GitHub Actions, Docker
- Engineering: FastAPI, Spring Boot, GraphQL, MongoDB, Kubernetes, AWS
- Michigan State University: B.S. Computer Engineering, Dec 2025 (GPA 3.31/4.0)
- Kalamazoo Valley Community College: A.S. Chemical Physics, Apr 2023 (GPA 3.94/4.0)