In this second issue of Scientifica Journal, we take you into the beating heart of innovation at Scientifica Venture Capital, the very hallmark of our identity: the Scientifica Lab, 1,400 square meters of proprietary laboratories located in L’Aquila, inside the Tecnopolo d’Abruzzo.

Guiding us through this journey is Simone Pesce, Science & Tech Project Manager, who will talk to us about artificial intelligence and predictive machine learning models.

Hi Simone, can you tell us what you’re currently working on?

The Scientifica Venture Capital team is currently developing a special tool designed to generate deal flow starting from a basic technology, idea, or patent.

Thanks to LLMs (Large Language Models) and NLP (Natural Language Processing), this tool can understand the technological and application context of an idea and identify related patents, theses, or scientific papers.

Very interesting! Can you explain more in detail what technologies it relies on?

The tool we’re developing is based on three specific technologies:

• NLP (Natural Language Processing): these AI algorithms are capable of analyzing, representing, and understanding natural language. This allows us to extract information more accurately from patents and scientific publications.

• Predictive Machine Learning Models: these models, built on advanced machine learning algorithms, are designed to make predictions or estimates about future events. By developing and refining these predictive models, which use both historical and real-time data, the tool will be able to anticipate market trends, emerging technologies, and potential innovation opportunities.

• Sentiment Analysis and Emotional Interpretation: through these NLP-based applications, we train the tool to understand text in a way that mirrors human comprehension, determining whether the emotional tone is positive, negative, or neutral. This will be useful in assessing market perceptions and emotional responses related to specific technologies or patents.

A very ambitious project. I imagine there have been some technical or scientific challenges during the development process. Can you tell us about the main ones?

Integrating different types of technologies has presented several hurdles.

One of the main challenges has been training the various AI models so that they effectively respond to specific needs.

For example, it’s essential that the tool is able to analyze and process texts in order to extract relevant information, such as NERs (Named Entity Recognition); or that it can accurately interpret scientific terminology and industry-specific expressions to deliver precise sentiment analysis of patents and publications.

Another major challenge in implementing AI for advanced research is the effective interpretation of natural language queries.

Users often formulate their searches using either colloquial language or highly technical jargon, which the platform must accurately interpret to return relevant and precise results. This requires a deep understanding of context, linguistic ambiguity, and semantic variation.

And what strategies are being used to overcome these challenges?

Maintaining the accuracy of AI models, such as those used in advanced research, sentiment analysis, and recommendation systems, is crucial to providing users with relevant and useful results.

This requires ongoing research to develop and fine-tune AI algorithms, with the goal of improving the accuracy of text analysis and information extraction. It also involves careful management of training data, selecting the appropriate algorithms, and continuously validating the models.

Thank you, Simone, for this fascinating journey inside the Scientifica Lab and into the world of artificial intelligence.

We’ll be back next month with another issue of Scientifica Journal.