EXPERIENCE

GYM BADGES EARNED ALONG THE WAY
BACK
BADGES

  1. I returned to BNY for a second internship, this time on the infrastructure side, and was selected as my team's lead. I coordinate our planning and own the cloud and network workstream, defining infrastructure as code with Terraform.

    WHAT I'M DOING

    • Leading task planning for the team and presenting our mid-term progress on project scope and architecture to engineering stakeholders.
    • Owning the cloud and network workstream, provisioning environments as code with Terraform.

    STACK

    TerraformCloudNetworking

    In progress — outcomes coming as the project ships.

  2. Selected for a ServiceNow externship providing exposure to modern development and infrastructure ecosystems.

    WHAT I DID

    • Developed proficiency in enterprise technology tools, platform workflows, and industry best practices through hands-on learning.

  3. When a service misbehaves, engineers lose time jumping between monitoring tools to find out why. Our team built a system that detects anomalies in real time and hands them to AI agents that suggest fixes and trace root causes. I owned the fault detection and the agent design.

    WHAT I BUILT

    • Anomaly detection: an Isolation Forest (300+ trees) trained on 4,000+ telemetry points across 8 features, flagging latency and Docker faults across 4 services and triggering the response pipeline.
    • A two-stage agent system: a first agent packages AppDynamics data to propose targeted fixes; if that fails, a 3-agent system uses Prometheus context and Git history to surface root causes and alternatives.
    • A Google Gemini fallback that queries incident and anomaly context from a PostgreSQL database.

    RESULTS

    • Resolved 500+ injected faults across fault-injection tests.
    • Presented the final demo and system design to 50+ BNY employees and fielded live questions on architecture tradeoffs.

    STACK

    Pythonscikit-learnAppDynamicsGrafanaPrometheusDockerGoogle GeminiPostgreSQLGitJira

  4. Glovebox labs depend on stable conditions, and problems are easier to prevent than to clean up. I built a digital twin that simulates the system's telemetry and flags sustained anomalies that signal short-term risk. The results went into a research report approved by the program director.

    WHAT I BUILT

    • A Python digital twin simulating 3,600+ timesteps of glovebox telemetry.
    • A feature selection process using regression analysis and variance inflation factor (VIF), reducing 8+ system variables to a statistically validated set.
    • An Isolation Forest model with temporal persistence to catch sustained anomalies rather than one-off spikes.

    RESULTS

    • 0.9 ROC-AUC for short-horizon risk prediction.

    STACK

    Pythonscikit-learnPandasNumPyMatplotlib

  5. Full fine-tuning updates every weight in a large language model, which is expensive. LoRA cuts the cost but doesn't say which parts of the model actually need to change. TopoTuner treats each attention projection matrix as a point cloud and tracks how its topological shape changes during fine-tuning, measured with persistent homology and Wasserstein distances. It learns which matrices to freeze, and that freezing profile transfers to new datasets.

    WHAT I DID

    • Co-authored the paper, contributing dataset curation, model evaluations and visualizations.
    • Built tooling to measure weight change across LLM attention components, analyzing 15+ freezing configurations over 100+ fine-tuning experiments on UCF's ARCC HPC cluster.
    • Delivered multiple 45–50 minute presentations on deep learning research papers to the lab.

    RESULTS

    • Competitive with full fine-tuning while training only 1–2% of parameters.
    • Outperforms LoRA in 7 of 9 model-dataset settings across LLaMA-3.1-8B, Mistral-7B-v0.3 and Qwen3-8B-Base.
    • Cuts training time by 20.4% versus full fine-tuning and 5.5% versus LoRA on average.

    STACK

    PythonPyTorchSLURMLLaMA-3.1-8BMistral-7BQwen3-8B

  6. Served as a mentor for 15 incoming freshmen in the Honors program majoring in Computer Science.

    WHAT I DID

    • Met individually with each student through June to advise on courses, the honors curriculum, and answer questions about navigating UCF.