From extensive experience in developing technologies for research, information processing, and AI, Istella’s proprietary language models are born, designed for conscious, controlled, and sovereign use of data, adaptable to different business contexts.
From applied research to AI engineering
Istella’s journey originates from consolidated expertise in developing technologies for search, information processing, language understanding, and machine learning. Over time, this experience has led to the design of AI models and solutions that are secure, customizable, and governable, engineered to operate in complex organizational contexts.
Istella makes its language models applicable in enterprise contexts by integrating them with data, workflows, and domain knowledge. Data is collected and organized, knowledge is structured and made accessible, language is analyzed and interpreted, while generative AI uses this context to produce responses, analyses, and operational support.
Orchestration and governance
The effectiveness of this approach depends on the ability to coordinate models, tools, data, and workflows in service of the organization’s objectives. Integration with corporate systems and customization of solutions enable AI to be adapted to different operational contexts.
Governance completes this framework through the definition of rules, access controls, monitoring, traceability, security, and compliance. In this way, the technology can operate reliably, transparently, and under the organization’s control.
The applied research pathway developed by Istella has given rise to a heritage of proprietary technologies, models, and patents, transforming methodological experimentation into solutions applicable to artificial intelligence, search, and information analysis.
Patents
Istella’s applied research is also reflected in its intellectual property. The company has developed proprietary technologies in the fields of machine learning and document ranking, transforming scientific and engineering expertise into solutions applicable to search and information analysis systems.
Among the publicly accessible patents assigned to Istella is the U.S. patent “Method to rank documents by a computer, using additive ensembles of regression trees and cache optimisation, and search engine using such a method”, which describes a machine learning-based methodology for ranking documents through decision trees and cache optimization techniques.
Scientific publications
Scientific publications document Istella’s contribution to applied research in the fields of ranking, machine learning, and evaluation of search systems. The presented works include datasets, algorithms, and methodologies developed to improve the efficiency, effectiveness, and scalability of systems that order and select information at scale.
- The Istella22 Dataset: Bridging Traditional and Neural Learning to Rank Evaluation — D. Dato, S. MacAvaney, F. M. Nardini, R. Perego, N. Tonellotto, SIGIR 2022 – This publication covers the dataset designed to compare traditional Learning to Rank approaches and neural models on the same information heritage.
- Fast Ranking with Additive Ensembles of Oblivious and Non-Oblivious Regression Trees — D. Dato, C. Lucchese, F. M. Nardini, S. Orlando, R. Perego, N. Tonellotto, R. Venturini, ACM Transactions on Information Systems, 2016 – This publication describes ranking methods based on ensembles of decision trees, in relation to the Istella LETOR dataset.
- Post-Learning Optimization of Tree Ensembles for Efficient Ranking — C. Lucchese, F. M. Nardini, S. Orlando, R. Perego, F. Silvestri, S. Trani, SIGIR 2016 – This work, developed by Istella researchers and collaborators, presents techniques for optimizing tree ensembles and making document ranking more efficient.
- Selective Gradient Boosting for Effective Learning to Rank — C. Lucchese, F. M. Nardini, R. Perego, S. Orlando, S. Trani, SIGIR 2018 – La pubblicazione, sviluppata da ricercatori Istella e collaboratori, descrive un metodo di gradient boosting selettivo per migliorare l’efficacia dei modelli di Learning to Rank.