מוצרים לחוקרים
Interactive sandboxes, experiment tools, model comparisons, reliability analysis, benchmarks, graphs, and research reports that shorten the path from a research question to a reproducible experiment.
Explore research productsמעבדת פרויקטי מו"פ
Project-based AI R&D for problems without a clear engineering answer: feasibility, baselines, model training and comparison, optimization, prototyping, code, models, and engineering reports.
Discuss an R&D projectAI Research Lab Projects
The lab is built for real research uncertainty: when it is not yet clear which approach will work, a previous project has stalled, or an algorithm or paper must become a working system.
We define the experiment, build the baseline, train and compare models, analyze errors, optimize results, and deliver a prototype with code, models, and an engineering report.
Software Engineering AI
Improving Requirements Classification
A machine-learning study using imbalance-aware preprocessing for requirements classification.
Read PaperSensor AI
Transformer Behavior Classification
A transformer encoder for accurate real-time classification from motion sensor data.
Read PaperRuntime Stability
Automatic Stability and Recovery
A runtime controller that detects destabilizing training updates and recovers by rollback.
Read PaperAgentic Reliability
MTTR-A Cognitive Recovery Latency
A reliability metric for measuring cognitive recovery latency in multi-agent AI systems.
Read PaperRAG Evaluation
From Likelihood Shifts to Defensible Claims
An audit of fixed-target likelihood measures that separates support shifts from preference, generation, and robustness claims.
Read PaperKalman Filtering
Selection-Induced Innovation Contraction
How validation gating and nearest-neighbor association reshape innovation statistics in Kalman tracking.
Read PaperPINN Optimization
The Direction-Execution Gap
An optimizer-aware audit for detecting when locally favorable PINN updates deteriorate after complete finite-step execution.
Read PaperSensor AI
Learning-Based Speed Estimation from Accelerometer-Only Inertial Sensing
A compact temporal model that estimates vehicle speed from smartphone acceleration alone and analyzes the role of finite sensing context.
Read Paper