Sanchit Sahay

Systems, Security, Cloud, Web and More!
HomeProjects & PublicationsPocket Litter
Publications
"Over the Shoulder: Improving SBOM Accuracy by Watching the Build" · PDF
Sanchit Sahay, Vyom Yadav, Abhishek Reddypalle, Uk Jang, Marco De Vincenzi, Santiago Torres-Arias, Justin CapposTo appear in Conference on Software Supply Chain Offensive Research and Ecosystem Defenses (SCORED '26), October 06, 2026, Prague, Czech Republic
ACM, New York, NY, USA · 11 Pages · 2026 · https://doi.org/10.1145/3848003.3848006
Research
  • SBOMit (OpenSSF)
    Maintainer / Researcher
    OpenSSF initiative to augment Software Bills of Materials (SBOMs) with build time in-toto attestations.
  • Lind-Wasm & 3i (Secure Systems Lab)
    Contributor / Researcher
    WebAssembly sandbox for isolating unmodified POSIX applications, with 3i routing system calls through composable userspace policies.
  • Open Source & Course Projects
    OPEN SOURCE
  • Witness (CNCF)
    Maintainer
    Witness is a pluggable framework for software supply chain risk management. It automates, normalizes, and verifies software artifact provenance.
  • Author
    A kernel module implementing Apple's HFS+ filesystem for FreeBSD 14 — modernizing the VFS-layer APIs, removing Darwin dependencies, and building the mount_hfs, newfs_hfs, and fsck_hfs userland utilities.
  • Contributor
    Contributor to Cargo, Rust's official package manager and build system.
  • A POSIX tail that supports grepping a log stream while it is still being written, via --follow.
  • A local stand-in for Google App Engine, so services can be run and tested without deploying to GCP. Grew out of the CI/CD work at LegalAI.
  • COURSE PROJECTS
  • Comparing Training Efficiency for Large Language Models
    Compares LLM fine-tuning strategies and documents the trade-offs between them: PEFT (LoRA, QLoRA, GoRA), quantization, and distributed-GPU training (DDP and FSDP).
  • Improving Learned Bloom Filters
    Benchmarks optimization techniques for Learned Bloom Filters, weighing false-positive rate against memory use across Projection Hashing, Caching, and LoRA.
  • A single place for college students to keep their academic and professional data. It pulls from Notion and Gmail to track job applications, and uses RAG over personal notes so they can be queried in plain language.
  • Predicts NYC subway ridership from temporal and fare-class features, and clusters stations to surface neighborhood-level traffic patterns.