Search results

Applied Scientist

Self-evolving skill libraries face a silent failure mode we term library drift: unbounded skill accumulation without outcome-driven lifecycle management causes retrieval degradation, false-positive injections, and performance stagnation. Recent evaluation confirms the symptom (LLM-authored skills deliver +0.0pp gain while human-curated ones deliver +16.2pp; SkillsBench (Li et al., 2026)), yet the underlying

In many time series forecasting settings, the target time series is accompanied by exogenous covariates, such as promotions and prices in retail demand; temperature in energy load; calendar and holiday indicators for traffic or sales; and grid load or fuel costs in electricity pricing. Ignoring these exogenous signals can substantially degrade forecasting accuracy, particularly when they drive spikes, discontinuities

Despite their strong zero-shot forecasting capabilities, Time Series Foundation Models (TSFMs) lack mechanisms for incorporating the structured domain knowledge that practitioners need for interpretability and forecast control. Dynamic Factor Models (DFMs) provide this structure, decomposing series into interpretable, adjustable factors like trend and seasonality while capturing shared dynamics across related

Data Scientist

Research Scientist

Video-Language Models (VidLMs) achieve strong benchmark scores, yet these scores often hide whether models use the video at all. We show that VidLM failures follow two pathways: some visual signals are never reliably encoded, while others are encoded but overridden by model priors. We introduce REVEAL, a diagnostic stress-test benchmark for quantifying when and why VidLMs under-use visual evidence. REVEAL

Container image pulling accounts for the majority of pod startup time in Kubernetes environments. Standard pull down loads the entire image before the container can start, even when the application accesses only a fraction of the image content at startup. We present SOCI (Seekable OCI), a lazy-loading architecture that enables containers to start without downloading the full image. SOCI builds an external

Evaluating rule compliance in industry requires assessing products against complex regulatory standards using multimodal data sources—a task where both correctness and trustworthiness of automated judgments are critical. Existing approaches either rely on costly human audits, supervised classifiers that demand large-scale labeled training data, or monolithic multimodal models that apply uniform reasoning

Recent advancement in vision-language models have enabled multi-modal person re-identification (Re-ID), where the system takes both an image and a text query to identify matching individuals. While previous state-of-the-art methods perform well with detailed, sentence-level descriptions, we found that their Recall@1 drops by half when using short, keyword-based queries due to ambiguity, training biases,

Evaluating multi-step diagnostic reasoning in LLM agents remains an open problem. When cause labels are extracted from resolved operational cases (customer-service tickets, incident reports, clinical notes), the resulting gold standards exhibit extreme vocabulary explosion—5,076 unique cause strings from 2,196 tickets on a single symptom, 92% appearing only once—making LLM-as-judge protocols variance-prone

Manager, Applied Science

At Amazon Prime Video, we face the critical operational challenge of managing code deployments during live events and rapid feature releases without causing service outages. Current change control approaches use blanket deployment freezes that block all changes regardless of risk, creating significant developer toil. While prior re-search has explored risky change predictors, these rely on developer-specific

Recent work identified Super Weights, individual parameters whose removal degrades model performance by orders of magnitude. We show that this degradation due to pruning Super Weights does not universally apply to all LLMs. Furthermore, if these parameters are so important, Super Weight-aware training should be effective. We show the opposite. Training Super Weights in isolation (100 to 8,192 parameters

Applied Scientist

Manager, Data Science

Applied Scientist

Applied Scientist

Applied Scientist

Data Scientist