Mohamed A M Elansary, PhD
Target: Member of Technical Staff — ML Research, Multimodal — Causal Labs
Sourced insights
- Multimodal physical obs → LPM architectures: Learn from sparse sensors, point clouds, hyperspectral imagery, physical fields at LLM-dwarfing scale; design architectures and training recipes that turn multimodal observations into a model that predicts the physical future — including encoding heterogeneous/irregular modalities, stable long-horizon rollouts, and probabilistic forecasting. Source: Ashby JD — MTS ML Research, Multimodal
- LPM learns causality from physics — not text/vision: Domains governed by physics have inherent cause-and-effect relationships, unlike visual or textual data — public Causal mission thesis for why an LPM starts from physical substrates. Source: causallabs.ai/mission
Proof — multimodal sensors × irregular sampling × long-horizon rollouts × probabilistic × ship
- PhD Environmental Engineering, TAMUK 2022: monthly-to-seasonal SW–GW forecast systems from heterogeneous USGS/NOAA/NASA physical fields — long-horizon physical prediction under uncertainty on HPC.
- Physical sensors → usable signals: GRA built environmental-monitoring from scratch (sensor selection, signal conditioning, data-loggers) — sparse / irregular modalities into pipelines.
- Physics-adjacent stacks (MODFLOW, VIC, PIHM, NASA LIS) + multimodel / probabilistic ensembles + AMS multimodel streamflow & floods/droughts — domain fluency for physical forecasting, not claimed LPM ownership.
- Research → scaled ship: Vertexium production agentic LLM + retrieval + multi-tenant isolation; Lucent CTO pipelines; HPC Linux 6+ years (UCAR, TACC).
- Honest frame: not Causal Eval framework owner. The PhD who ships. SF onsite OK · Prefer take-home.