Refer to github.com/kasra-hosseini for all open-source projects.

Selected Projects

• MapReader
  An end-to-end computer-vision pipeline for the semantic exploration and analysis of images at scale. Applied transfer learning on pretrained vision models (CNNs, vision transformers) to classify ~16k historical maps (~30M geospatial patches). Winner of the Roy Rosenzweig Prize; adopted externally for plant phenotyping.
  SIGSPATIAL · JOSS · DH Awards 1st Runner-Up
• DeezyMatch
  A flexible deep learning approach to fuzzy string matching and candidate ranking using deep learning and contrastive learning.
  EMNLP 2020
• histLM
  Neural language models (word2vec, fastText, BERT, Flair) for historical research, trained on ~5B tokens of nineteenth-century English. Models available on Hugging Face and Zenodo.
  JOHD 2021
• scivision
  A toolkit for scientific image analysis, connecting computer vision models to scientific imagery. Contributor.
• obspyDMT
  A Python toolbox for retrieving, processing, and managing large seismological datasets.
  Solid Earth 2017
• SubMachine · homepage
  Web-based tools for exploring 3-D seismic tomography and other models of Earth's deep interior. 750k+ queries served as of March 2026.
  G³ 2018 · Front Cover Image · Top 20 most read
• privgem
  Privacy-preserving generative models (GANs) for synthetic data generation.
• GeoTree
  Nearest-neighbour searches on geographic coordinates using KDTree and BallTree.

Contributions

QUIPP-pipeline (privacy-preserving synthetic data generation) · PressPicker (interactive visualisation for picking newspaper titles) · daedalus (dynamic spatial microsimulation pipeline for population projections) · AxiSEM (spectral-element wave propagation) · obspy (Python framework for seismological data) · MC Kernel (sensitivity kernels via Monte-Carlo integration)