Agentic AI for R&D Acceleration
Research and development of agentic AI systems capable of automating and accelerating scientific research workflows: literature review, experimental design, data analysis, and writing.
The problem
Scientific research faces a growing bottleneck: the volume of published literature far exceeds human processing capacity. A researcher spends weeks on literature review, months on data collection and cleaning, and a significant proportion of their time on repetitive tasks that, while necessary, do not require creativity or expert judgment. Agentic AI can transform this landscape.
What is Agentic AI?
Agentic AI systems are autonomous systems capable of planning, executing, and adapting sequences of actions to achieve a complex goal. Unlike a language model that answers a question, an agent can decompose a task into subtasks, use external tools (academic search engines, databases, code environments), evaluate its own results, and course-correct when needed. In the R&D context, this means an agent can conduct a complete literature review, run statistical analyses, or draft sections of a paper—all with human oversight but without constant manual intervention.
Research lines
Automated literature review
Agents that search, filter, summarize, and synthesize relevant scientific publications for a research question. Capable of operating on academic APIs (Semantic Scholar, OpenAlex) and generating structured reports.
Experimental design and execution
Systems that propose experimental designs, write and execute analysis code, and present results reproducibly. Integration with scientific computing environments (Python, R, Jupyter).
Scientific writing assistance
Agents that generate section drafts, verify consistency with cited literature, check target publication formatting, and suggest improvements in style and structure.
Workflow orchestration
Orchestration frameworks that coordinate multiple specialized agents to complete complex research tasks end-to-end, with human checkpoints.
Our approach
Responsible AI
Every agentic system we develop includes human oversight mechanisms, decision traceability, and risk assessment. The goal is not to replace the researcher, but to amplify their capacity.
Measurable impact
We measure real acceleration: time saved on literature reviews, error reduction in analyses, quality of generated drafts vs. manual writing. Reproducible results published openly.
Open science
Tools, datasets, and experimental results published under open licenses. Any research group can replicate and adapt our systems to their own needs.
Applications
- Agent-assisted systematic reviews (systematic reviews in hours, not months)
- Automated bibliometric analysis and scientific field mapping
- Hypothesis generation and verification from existing literature
- Automated preparation of funding applications and project proposals
- Continuous monitoring of new publications relevant to active research lines
Publications
openreview-mcp: A Model Context Protocol Server for Querying Peer Review at Scale
F.J. Rodrigo-Ginés — OpenCódice Technical Report OC-TR-2026-007, 2026
The Agentic R&D Decalogue: Ten Principles for Human-Agent Collaboration in Scientific Research
F.J. Rodrigo-Ginés — OpenCódice Working Paper OC-WP-2026-001, 2026
The Transparency Card: A Structured Reporting Framework for Agent-Assisted Research
F.J. Rodrigo-Ginés — OpenCódice Technical Report OC-TR-2026-006, 2026
Current status
The project is in active development. The first publications are now available as open-access papers on Zenodo.