From spec sheets to text and video, and a faster website
Technobattles.com is a website about technology products. I worked on it on two fronts: research on generating product descriptions automatically and, more recently, an overhaul of the site’s performance.
Descriptions generated from data
Every product has a structured sheet with dozens of specifications, and writing a readable description for each one by hand does not scale. With the research group at Politecnico di Torino we designed a data-to-text and text-to-speech pipeline based on open-source pre-trained language models and NLP techniques such as Named Entity Recognition: it starts from the spec sheet and produces a complete text description and a video version with a synthetic voice.
The goal was business-grade quality without relying on the most expensive models, which are also the hardest to integrate. The work was presented at IEEE AICT 2023 and SEBD 2024.
The performance overhaul
More recently we reviewed the site to improve its performance, working on the code with an AI-assisted development workflow (Claude Code): finding the bottlenecks, making targeted changes and checking the results.
Scientific publications
- A. Avignone, A. Fiori, S. Chiusano, G. Rizzo. Generation of textual/video descriptions for technological products based on structured data. 2023 IEEE 17th International Conference on Application of Information and Communication Technologies (AICT), pp. 1–7, 2023. Read →
- A. Avignone, A. Fiori, S. Chiusano, G. Rizzo. From Product Sheet to Text and Video: A NLG Pipeline to Transform Structured Data into Comprehensive Descriptions. Proceedings of the 32nd Symposium of Advanced Database Systems (SEBD 2024), CEUR Workshop Proceedings, pp. 271–280, 2024. Read →
AI for SMEs
Do you have catalogues or product sheets to turn into content? Language models can do it well, if the pipeline is designed around your data.