Writing a Marie Skłodowska-Curie Postdoctoral Fellowship proposal means balancing Excellence, Impact and Implementation. These three aspects must align with the call text, the expectations of the evaluators and the research idea of the candidate, all while respecting the strict page limits and deadlines.

ASTRA was conceived, within the IANUS project, to help the applicants to meet this challenge. ASTRA is not only a writing assistant; it is a tool for the development and the self-evaluation of the project proposals based on artificial intelligence and designed specifically for MSCA Postdoctoral Fellowships, fine-tuned on a dataset containing real project proposals and their respective evaluations.

ASTRA is designed to support researchers during the whole life cycle of the proposal, from the preliminary ideas to the final refinements. The tool is based on the assumption that successful proposals have similarities that can be found in the structure, in the language and in the strategic choices. Patterns that can be analyzed, learned and re-used.

ASTRA is not a substitute for the researcher in the proposal writing, but it acts as a guide that helps structuring ideas, formalizing the language and identify weaknesses when there is still room and time for improvement.

 

Three functionalities to support MSCA applicants

ASTRA has three core functionalities that are strictly connected, each designed to face one of the key challenges in the MSCA proposal writing process.

1. Guided proposal drafting: ASTRA assists users section by section, providing contextualized suggestions, based on the schemes observed in the real proposals used for the fine-tuning. This way, ASTRA helps applicants turning their ideas into texts that are coherent with the structure and the expectations of the call. This is particularly important for researchers that are not familiar with project proposal writing. People who may have great research ideas, but who are penalized by their limited experience with the formal language of European founding.
2. Draft refinement and self-assessment: Another useful functionality offered by ASTRA is the ability to act as a critic. The users can upload whole sections – Excellence, Impact or Implementation – and receive a feedback aligned with the real ones of the Evaluation Summary Reports (ESRs). This tool highlights the most common errors, signals misalignments with evaluation criteria and generates operative recommendations. The goal is not to emulate an evaluator, but to help applicants to understand how evaluators read and judge proposals.
3. Keywords and summary generation: Starting from the description of an idea, a schema or a complete proposal draft, ASTRA can generate an abstract, a keyword list or a structured summary. These outputs are aligned with the language and the structure learned from the real examples seen during the fine-tuning phase.

 

How does ASTRA work?

ASTRA is a modular AI system that combines language models, real and structured knowledge sources, and workflow automation. Instead of solely relying on a generic training, ASTRA is based on an LLM (Mistral) that was fine-tuned using a dataset that contains real project proposals and their corresponding evaluations, from which all sensitive data were removed in advance.

The drafts uploaded to the model from the users are also sensitive documents, therefore ASTRA has been designed with GDPR compliance and data security as fundamental principles. The system can be deployed using optimized language models within secure cloud environments, ensuring that user data are neither disclosed nor stored in the cloud. Session-based memory allows ASTRA to maintain context during multi-step workflows, without building long-term user profiles or retaining data beyond the strictly necessary time.

ASTRA does not involve any form of continuous training, which means that users’ conversations and any uploaded files will not be used for further training of the model and will therefore remain solely in the possession of the researchers.

 

A tool, not a shortcut

ASTRA does not promise guaranteed success and does not automatically write complete proposals. On the contrary, it supports a more informed decision process, clearer writing and early identification of critical issues.

ASTRA was designed, within the IANUS project, as a concrete example of how AI can be used in a responsible and strategic way to help researchers, without compromising academic integrity and creativity.

In an increasingly competitive funding landscape, tools like ASTRA point to a possible future direction, in which researchers spend less time interpreting formats and expectations, and more time refining their research ideas.