SWE @ Microsoft | Ex-MLE @ Mercer|Mettl
I build things at the intersection of infrastructure and intelligence -- and random stuff too.
Currently engineering Azure UltraDisk at Microsoft. Before that, shipped ML systems at Mercer|Mettl mostly around speech detection across regions and accents.
- Mainly Ultradisk, which is the Azure's premium high performance offering for IO-sensitive, latency-sensitive workloads.
- Tinkering and Maintaining growing collection of random projects and research.
Scholar MCP -- Local MCP server that lets AI coding agents search Google Scholar for papers by topic or author.
Bookworm -- Multi-modal book recommendation engine using semantic search across plots, authors, and genres.
Projects I've built and documented. Old Stuff
ML Ransomware Detection -- Catching Ransomware in Virtual Machines (NetApp)
A two-phase ML detection system built for VMware environments. Monitors VM snapshot disk files for anomalies, then drills into file structure using heuristic and signature-based methods.
Stack: Python, scikit-learn, VMware VMDK parsing
How it works:
- Phase 1: Anomaly detection on VMDK snapshot files -- catches unusual patterns before encryption completes
- Phase 2: Deep file structure analysis with heuristic + signature-based methods
- Parses raw VMDK formats into feature-extractable representations
- Real-time monitoring capable of detecting ransomware as it happens
Built during my time working with NetApp. The code is proprietary, but the methodology is reproducible. Designed for cloud environments where ransomware targets virtual machine disk storage.
Scientific Paper Data Extraction -- Automated Time Series Pipeline from arXiv
End-to-end pipeline that downloads scientific papers from arXiv, extracts figures, identifies line charts, and converts them into structured time series data. Minimal manual intervention.
Stack: Python, PDFFigures2, SVG processing
Pipeline stages:
downloading_arxiv.py-- Bulk paper fetch from arXivfiltering_pdfs.py-- Relevance filtering- PDFFigures2 -- Figure and caption extraction
crop2pdf.py-- Precise document croppingisLine.py-- Line chart classificationpdf2svg.py-- Format conversionsvg2time_series.py-- Final data extraction
Researchers spend hours manually pulling data points from published charts. This pipeline does it automatically at scale.
This profile is a living document. The projects folder grows as I build.