As part of the admissions process for the Scientific and Technological Researcher Career (CICyT) of CONICET, whose evaluation committees began their work in June, the organization implemented an artificial intelligence (AI)-based system for the identification and selection of specialists. This development is a joint effort of the Human Resources, Evaluation and Planning, and Organization and Systems Management departments, in collaboration with staff from the Higher Institute of Software Engineering (ISISTAN, CONICET-UNCPBA), which belongs to the CONICET Tandil Scientific and Technological Center.
As with other modernization initiatives within the organization, this tool does not replace the judgment of advisory committee members, the CONICET stated. Its function is to expedite the preliminary search for potential evaluators, eliminating the manual task of tracing profiles and providing technical input that optimizes the specialist assignment process carried out by the advisory committees.
This tool suggests potential specialists by automatically cross-referencing millions of data points on scientific and technological output. In addition to streamlining this process, the system can identify specialists from outside the applicant’s discipline or committee of origin. This is particularly relevant for evaluating projects with multi- and transdisciplinary, technological, or applied approaches, which often present a challenge for advisory committees structured around specific disciplines.
This innovation is part of a CONICET initiative launched in 2025, whose objective is the gradual integration of AI to optimize institutional management.
Technical operation of the specialist recommendation system #
The recommendation tool relies on three CONICET systems: the Institutional Repository (RI), the Comprehensive Science and Technology Search Engine (BICYT), and the Peer Consultant Database. The recommendation process compares the applicant’s work plan and background information recorded in the Comprehensive Management and Evaluation System (SIGEVA) with the track record of experts in the peer consultant database registered in SIGEVA, through a sequence of processing and filtering stages:
-
Semantic similarity analysis: The work plan and scientific-technological outputs (titles, abstracts, and keywords) are analyzed using natural language processing techniques. This allows the system to determine the conceptual meaning of the information and identify the affinity between the applicant’s profile and the evaluators, even if they use different terms or are in different languages.
-
Automatic detection of conflicts of interest: the system applies rules to exclude potential evaluators with incompatibilities with the applicant. Supervisors or co-supervisors of each applicant, researchers belonging to the same workplace (according to their identifier code) where the applicant is applying, co-authors of publications, and those with the same or lower research rank than the one requested by the applicant are automatically excluded.
-
Two-stage affinity processing:
-
First stage (general search): performs a quick comparison of semantic similarity between the applicant’s profile and the output of the bank’s specialists.
-
Second stage (deep refinement): the highest-scoring pre-selected potential evaluators are analyzed with a more complex language model (based on deep learning neural networks), which performs a more thorough examination of the text of the publications to determine the degree of correspondence more accurately.
Disciplinary alignment: To balance scientific specialization with openness to other areas, the system weights suggested specialists according to their belonging to a larger area or discipline different from that of the applicant, ordering the suggested recommendations more precisely.
Following the analysis, the members of the advisory committees will have a set of specialists for each application, considering the best match according to the applicant’s work plan and their Science and Technology production.
Contact [Notaspampeanas](mailto: notaspampeanas@gmail.com)