Bỏ qua đến nội dung chính
Back to home
AI tools-ai 1 min read

Introducing DeepTS and DeepScribe: Two AI agent frameworks for scientific research

Researchers have introduced DeepTS and DeepScribe, two autonomous AI agent frameworks designed to automate time-series data collection and convert complex physics lectures into structured scientific reports.

Tier 2 · sources 99% confidence Reviewed
Sources arxiv.org

A research team has recently announced two new frameworks designed to develop autonomous AI agents for scientific research workflows, utilizing a hybrid architecture that combines local orchestration with cloud processing.

Key Developments

The system is based on a 'Local Body, Remote Brain' architecture via Google Colab, employing local Python orchestrators to call large language models (LLMs) in the cloud. The first framework, DeepTS/DeepCollector, specializes in automating the collection, extraction, and deduplication of large-scale time-series datasets. The second framework, DeepScribe, is an autonomous presentation analyzer capable of converting visually dense and mathematically complex physics lectures into structured scientific reports.

To achieve this, the research team applied practical systems engineering techniques such as fine-grained attribute extraction (Cellular RAG), remote data validation, and distributed concurrency control. This approach helps AI agents overcome the context and reasoning limitations of current systems, providing robust support for scientific workflows.

Why It Matters

The application of agentic AI in science marks a shift from using AI as an auxiliary tool to employing systems that autonomously run research workflows. For the scientific community in Vietnam, open-source frameworks like these can help ease the burden of raw data processing, allowing researchers to focus on hypothesis generation and in-depth analysis. Furthermore, future expansions into knowledge graphs and high-energy physics (DeepQCD) promise to open up new application spaces for AI in fundamental research.