ML systems, geospatial intelligence, document AI
An Applied Research Scientist adapting computational intelligence to ML/AI systems & real-world messy data:neural retrieval, document intelligence, geospatial imaging, and scientific data infrastructure.
I am curious and enthusiastic about SYSTEMS: ML Computing Systems, Data Systems, Knowledge Production Systems, Governance Systems.
About me / How I can help
My daily tasks involve cross-functional teams collaborations to create and deliver unprecidented approaches towards unyielding industrial challenges with application of computing, and more - from research framing through prototype, evaluation, and production handoff. My current work and interests include;
I prototype and evaluate extraction pipelines for mixed-format document corpora.
I design table and layout understanding workflows for noisy, heterogeneous reports.
I help teams productionize ingestion, validation, and downstream integration for document intelligence systems.
I develop segmentation and imaging workflows for material, mapping, and scientific-data problems.
I design geospatial ML and computer vision architectures for earth systems, infrastructure, and resource assessment work.
I build benchmarks that reflect operational limits instead of clean-room assumptions.
I translate mathematical concepts into algorithmic systems and testable engineering roadmaps.
I design scalable ETL, model development, and inference strategies for teams working with large or irregular datasets.
I work with product, research, and engineering teams to move from open-ended questions to measurable systems.
My Previous Record
A decomposed record of the work: what I am studying, what I have published or built, and the roles that shaped the current research direction.
2026
Thesis
This thesis stress-tests embedding retrieval architectures across controlled degradation conditions: baseline, bandwidth constraints, high latency, and packet loss. The work compares how HNSW and IVF fail differently, and treats recall, latency, and energy per query as linked system outcomes.
2025
Preprint
Co-authored work reframing binary classifier evaluation as a numerical conditioning problem. The project connects ROC, PR, and F1 behavior on a differentiable surface and studies which bounded optimization routines produce stable thresholds under noisy empirical metrics.
2024
Product
A document-intelligence utility for recovering structured tables from reports that vary in layout, encoding, and schema. The core problem is disambiguation: merged cells, multi-header tables, rotated layouts, and documents that were never designed for machine reading.
Ongoing
Notes
A chapter-by-chapter study of parallel computing applied to machine learning systems and large-scale data structures. The separate Parallel Systems page keeps its own terminal-inspired IBM Plex and Syne style.
Open Parallel Systems Notes2023 - present
Embedding knowledge systems and computer vision for mineral regimes, with mathematical models and pipelines that reconstruct fragmented oil-and-gas records into usable national science assets.
2024 - 2026
Computing systems and algorithms research focused on retrieval architecture resilience, parallel systems, benchmarking, and realistic network conditions.
2021 - 2022
Research on ML training data and large-data system design, including losses created by schema variation in open-source data.
2017 - 2023
Expanded open-science curriculum for GIS and ML application, created open geospatial datasets, and supported a global community of students, practitioners, and researchers.
2018 - 2020
Mapped and advised schema variations for land-data mapping and national record digitisation, with work spanning data management, processing, and software development.
PDF
PPTX
The full resume includes roles, publications, projects, and technical skills in a downloadable PDF. The summary deck offers a quick PowerPoint overview of the same record.
Section 3
Away from research, I keep a small room for books, philosophy, historical fiction, films, and the story worlds that help me think more slowly.
Enter the Reading Room