ELIS: A Multi-Agent AI Platform for Scientific Research and Engineering

The ELIS multi-agent AI platform aims to address the time-consuming and labor-intensive pain points of Systematic Literature Reviews (SLR) in academic research and engineering practice. It provides a one-stop solution for data collection, deduplication, screening, validation, and auditing through a unified CLI tool and reproducible pipelines, supporting reproducible research and engineering workflows.

In academic research and engineering practice, SLR requires collecting literature from multiple databases, merging and deduplicating, screening and validating, etc., which is a tedious process. ELIS initially started as a proxy repository focused on SLR, then evolved into a multi-agent platform supporting reproducible research and engineering workflows. The project package namespace and CLI still retain the name elis, and the current active branch is release/2.0; version 2.0.0 marks its entry into a mature and unified phase.

The core design of ELIS revolves around four principles: 1. Unified CLI interface integrating full-process operations; 2. Reproducible pipelines: outputs of each run cycle are stored in runs/<run_id>/, including intermediate products and manifests; 3. Staged deterministic outputs: products of each stage (harvest→merge→dedup→screen→validate) are transmitted in JSON format; 4. Audit and traceability: run manifests must comply with predefined JSON Schema to ensure verifiable steps.

The CLI covers full-process commands: data collection supports OpenAlex, Crossref, Scopus, etc. (Web of Science and others are planned to be added); merging multi-source data; deduplication to identify duplicate literature; screening based on inclusion/exclusion criteria; validating data formats; exporting results. It also provides agentic sidecar workflows, such as elis agentic asta discover and enrich, to assist in literature discovery and metadata enrichment.

The project has a comprehensive documentation system (e.g., RELEASE_PLAN_v2.0.md) to ensure sustainability. Technically, it is developed in Python, and deployment recommends using a virtual environment: create a venv → activate it → install dependencies, and view command help via elis --help.

Applicable scenarios: standardized SLR processes, bibliometric analysis, research reproducibility, multi-source data integration. Summary: ELIS combines multi-agent AI with literature review workflows to improve efficiency and standardization. Outlook: More database adapters will be added, AI agent capabilities will be enhanced, and it will become an important infrastructure for academic research.