← Back to Home

Experience

Edgescale AI

Software Engineer, AI Agents

Edgescale AI • Denver, CO

May 2026 - Aug 2026

  • Built an internal agent platform with container admission gating, isolated Kubernetes runtimes, and an Okta-gated console. Runs 6 production agents on 3 edge clusters.
  • Built a code-review agent: 15 review skills across 7 specialist lanes, running in a deterministic LangGraph kernel with generator/judge verification.
  • Cut the reviewer's false positives 73.5% (427 to 113) across 659 verified findings, at about $3 per review.
  • Shipped a gated self-improvement loop that mines the fixes for shipped defects into new review rules.
  • Built a self-improving skill loop on edge GPUs. 13 parallel lanes write skills and verify them against robot simulation, emulator state, and reproductions of 9 ML papers, keeping only results that clear a 95% statistical bound.
LangGraphMCPKubernetesDockerTerraformAWS
Linot Lab, UMass Amherst

Research Assistant

Linot Lab, UMass Amherst • Amherst, MA

Jan 2026 - Present

  • Reconstructing vorticity and velocity fields from sparse streakline images for inverse problems in fluid dynamics.
  • Designing and testing CNN + MLP/PCA models in PyTorch on the UMass Unity HPC cluster.
  • Synthesized a 50K-sample Kolmogorov flow dataset to measure reconstruction accuracy, generalization, and whether multimodal inputs are feasible.
PyTorchCUDASLURM
Nursing & Engineering Lab

Research Software Engineer

Nursing & Engineering Lab • Amherst, MA

Sep 2025 - Dec 2025

  • Found and fixed logic and numeric errors in an embedded EMG signal processing pipeline.
  • Fixed wrong variance and RMS values by repairing buffer indexing and the statistics.
  • Cut runtime by ~50% with buffer management and control-flow changes, enough to run on the embedded device.
CEmbedded SystemsSignal Processing
SciQuel

Software Engineer Intern

SciQuel • Remote

May 2025 - Sep 2025

  • Built REST APIs in TypeScript, Next.js, and React that post and fetch scientific article contributors from MongoDB and render them on article pages.
  • Redesigned the Prisma schema so scientific contributors are separate from site users, and kept the existing APIs usable.
Next.jsTypeScriptPrismaTailwind
Forschungszentrum Jülich

Machine Learning Engineer Intern

Forschungszentrum Jülich • Jülich, Germany

May 2025 - Aug 2025

  • Fine-tuned DINOv2, a self-supervised vision transformer, on multi-terabyte hyperspectral imagery datasets, running distributed jobs on HPC through SLURM.
  • Cut environment deployment time 70% by automating infrastructure orchestration and resource allocation with Docker and SLURM job arrays.
  • Raised distributed processing throughput 30% with better data handling and caching across multi-node training jobs.
  • Supported by the DAAD RISE Scholarship.
PythonPyTorch LightningMLflowDockerSLURM
Build UMass

Software Engineer

Build UMass • Amherst, MA

Feb 2025 - May 2025

  • Built and deployed a chatbot for NEFAC that answers support questions.
  • Improved the FastAPI and GraphQL backend, cutting API response times by 40% and handling 20+ concurrent queries per second. The API connects to a React frontend.
  • Added memory and context retrieval with FAISS and LangChain so answers can use the earlier conversation.
TypeScriptFAISSLangChainFastAPI
Salesforce

AI / ML Fellow

Salesforce • Amherst, MA

Aug 2024 - Dec 2024

  • Built Random Forest models for SaaS customer sentiment and lead conversion prediction, at 91% accuracy.
  • Used Prophet to forecast when to reach out.
PythonPandasProphetSQL
Tufts University

Research Assistant

Tufts University • Medford, MA

May 2024 - Aug 2024

  • Worked with Prof. James Murphy on hyperspectral image clustering and unmixing.
  • Supported by the Tufts VERSE REU.
PyTorchPythonMATLAB