Customer-obsessed science
Research areas
-
November 16, 202611 min readNew open-source tool turns existing test suites into sets of formal-verification tasks, which catch buggy solutions that tests alone would miss.
-
November 11, 20265 min read
-
-
-
Featured news
-
Production LLM agents often run the same logic—including tool-calling loops, retrieval chains, and compositional workflows—in several operational modes. However, execution semantics are usually coupled to one runtime. We encountered this portability problem in Rufus, a conversational AI assistant that integrates a large catalog of tool endpoints and serves millions of Amazon customers [6]. Systems such
-
2026With Large Language Model (LLM) pre-training and fine-tuning shifting its focus from data volume to data quality, quality data selection has emerged as a critical research topic. Existing online data selection methods for LLM training are typically 'batch-constrained', limiting optimization to local utility within random batches. To overcome this, we propose GAIA (Global Adaptive Instruction tuning via
-
International Workshop on Acoustic Signal Enhancement (IWAENC) 20262026We present a real-time system for computing user location and orientation under constraints of memory, computation, and latency. The system utilizes a set of distributed microphone arrays with independent clocks. It computes angle-of-arrival (AoA) and time-difference-of-arrival (TDoA) information asynchronously, and deploys late fusion to compute the maximum likelihood user position. The user orientation
-
EMNLP 2026 Workshop for Research on Agent Language Models (REALM)2026The reliability of Large Language Models (LLMs) for API calling degrades in multilingual settings. A common failure occurs when a model selects the correct tool but generates argument values in an inconsistent language, which we term Argument Language Mismatch (ALM). Although semantically correct, such outputs are operationally invalid and not captured by standard API-calling metrics. We revisit post-training
-
2026Local field potentials (LFPs) capture neural population activity across brain regions and behavioral states. Learning transferable representations of LFPs could make it easier to compare recordings across experiments and support downstream analyses when labeled data are limited. Although many recent neural foundation models have focused on spiking activity, learning representations that generalize across
Collaborations
View allWhether you're a faculty member or student, there are number of ways you can engage with Amazon.
View all