Sep 2026· Workshop Proceedings of the 55th International Conference on Parallel Processing· pp. 133-136· 0 citations· 1 references
TL;DR
BriskSnapshot is a live demonstration of a unified semantic-window system implemented over a streaming AI control layer and a vector-stream runtime that follows public vulnerability-intelligence reports as they are normalized, embedded, joined inside a bounded semantic window, and surfaced as snapshot contracts for a provider-agnostic generator service.
Abstract
Streaming AI applications need to keep vector state fresh while exposing that state in a form that downstream services can inspect and act on. BriskSnapshot is a live demonstration of a unified semantic-window system implemented over a streaming AI control layer and a vector-stream runtime [7]. The demo follows public vulnerability-intelligence reports as they are normalized, embedded, joined inside a bounded semantic window, and surfaced as snapshot contracts for a provider-agnostic generator service. Unlike a static retrieval demo, BriskSnapshot lets the audience configure the stream, join strategy, retention window, parallelism, and generator backend while seeing how those choices affect runtime evidence and downstream answers. The live walkthrough stays small enough for narration, while a 3,000-record NVD replay and a 20-answer generator run show measured behavior across the vector runtime and contract-to-generation path.
AI auto-research is poised to forge bespoke data systems, from key-value stores and indexes to query engines, each engineered for a specific workload and hardware platform. Yet this promise rests on fragile ground. An agent optimizes what it can measure, not necessarily what users need. Natural-language requirements le...
Kai-Wen Zheng, Srijan Srivastava, Zhen-Feng Yan et al.· 0 citations
Modern teams depend on PostgreSQL for mission-critical workloads, yet administration is still done by hand-writing SQL in disconnected tools such as psql, pgAdmin and DBeaver, while AI assistants such as GitHub Copilot have no live connection to the database and cannot read its schema, execution plans or error output....
Shruti Mahendra Raskar, Brijendra Gupta· Iconic research and engineer...· 0 citations
Dynamic layer skipping reduces LLM computation by allowing each token to execute only a subset of the model's layers. However, existing skippers rely on specialized generation loops and do not integrate with modern serving engines. As a result, fewer executed layers do not necessarily translate into lower serving laten...
Wei Da, Yavuz Ferhatosmanoglu, E. Kalyvianaki· arXiv.org· 0 citations
Edge SoCs combine CPUs, GPUs, and NPUs, yet runtimes for multimodal large language models (MLLMs) keep generation on one engine and leave others idle. Streaming workloads allow ingestion and answering to overlap, but backend-local KV state binds queries to the engine that built their history. Stage performance and co-r...
Yi-Long Li, Cheng-Po Yan, Yi-Jing Zeng et al.· 0 citations
Understanding user behavior in extended reality (XR) is often performed through post-hoc pipelines that log complete sessions and invoke large language models (LLMs) offline. Such pipelines incur substantial latency, bandwidth overhead, and unnecessary LLM inference costs, limiting continuous in session analytics. We p...
Bo-Yan Li, Chong-Li Zhang, Chun-Sheng Wang et al.· IEEE Transactions on Visuali...· 0 citations
The reimplementation of single-execution grading certifies 86-100% of the pipelines eleven models produce; re-executing the same artifacts finds 7.0-79.2% of the certified ones silently wrong.
Pranay Kothari· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.