Personal Home Page / Est. 2026
Engineering & Growth · Incredibuild / islo.dev
Engineering / Research / Teaching
At Incredibuild / islo.dev, I work on engineering and growth for AI infrastructure. My background spans Linux kernels, computational genomics, and machine learning for autonomous driving. I have a PhD in physics and teach software engineering at HIT.
150 merged pull requests in 35 open-source repositories, 112 of them in 2026. See all on GitHub
Added the SandboxToolset that lets AI agents run shell and file tools in disposable sandboxes, then backends for Islo, Boat, OpenSandbox and OpenShell, and CI speedups.
Added 18 sandbox providers, including E2B, Modal, RunPod, Google Cloud Run, Azure Dynamic Sessions and Windows Sandbox, plus a provider-neutral heartbeat that keeps a delegated lease alive.
Replaced an arbitrary dispatcher-class hook with Spark-owned routing from the UDF worker specification, so operator-provided JVM code never loads inside Spark.
Made the Community Review Labeler read every page of a PR's reviews and fail closed on partial output, so PRs with more than 30 reviews are labeled from the latest trusted review.
Fixed a Linux thread-lifetime bug that killed sandbox helpers mid-request and caused intermittent file-edit failures.
Made sccache entries reusable across CI checkout roots by normalizing paths with SCCACHE_BASEDIRS and upgrading sccache to 0.16.0.
Isolated, auditable workspaces for concurrent coding agents: when two agents publish over each other, the second one's work survives as a fork.
GitHub issues to reviewed pull requests with coding agents, Airflow, and human approval gates; every run keeps its evidence.
GRCh38 vs the telomere-to-telomere T2T-CHM13 human reference, chromosome by chromosome: 179.6 Mb of added usable sequence and where CpG-rich windows fall.
Mobileye
Graphs of AI Coding: Orchestrating Agentic MapReduce with Airflow's New Sandboxes
MCP Is Not a Security Boundary
Your Agent Escaped Without Escaping the Sandbox
Forkable Sandboxes: The Runtime Layer for AI Software Factories
Nat. Commun., PNAS, Phys. Rev. E, ENCODE · Google Scholar
Acquired
Principal Engineer, AI Systems
AI execution infrastructure spanning forkable Linux sandboxes, CI runners, snapshots, caching, traces, reproducible environments, and build acceleration; public work includes agentic MapReduce, OSS CI experiments, and technical writing on Bazel, GitHub Actions, Rust, and kernel builds.
Research Scientist & Visiting Scientist
Computational biology and scientific pipelines connecting cancer transcriptomics, genomics, and quantitative computation, including patient-specific malignant-cell identity in tumor transcriptomes.
Tech Lead, ML/AI
Led ML and AI work across product and engineering, translating model capabilities into production systems and product decisions.
Deep Learning & Algorithms Researcher
Production perception and road-junction semantics for autonomous driving, with emphasis on reliable global scene understanding and difficult rare cases.
ML & Data Scientist
Industrial genomics and machine learning: genome analysis, CRISPR prediction, iterative learning, and cloud infrastructure for gene-editing workflows.
Computational Scientist
Computational genomics, 3D genome architecture, Hi-C, ENCODE-scale pipelines, physical biology, simulations, graph theory, and machine learning.
Software, firmware, kernel, and algorithms roles
Systems foundations across container security and firmware; at Marvell, Android/Linux-kernel work on PXA mobile platforms used by Samsung devices; at BAE Systems / Rokar, GPS and navigation algorithms in a guidance-systems environment.
University of Houston / 2020
The Hebrew University of Jerusalem / 2015
The Hebrew University of Jerusalem / 2014
The Hebrew University of Jerusalem / 2012
Senior Lecturer & course coordinator — Engineering of Advanced Software Solutions (EASS) / 2022 — Present
Lecturer / 2021 — 2023
Lecturer and mentor / 2016 — 2020
Teaching Assistant / 2010 — 2012
Short notes on AI systems, infrastructure, research, and teaching.