Markdown Price Optimization
Definition of a price recommendation system to be assigned to clothing items in the following sales season, in order to maximize profits.
YEAR 2011
CUSTOMER Fashion
METODOLOGIA
- Supervised learning techniques (boosting) for predicting sales at a given sale price
- Choice of price that maximizes profit
- Explainable AI techniques for evaluating reliability and explaining model results.
Classification of fashion products
Clustering of models on sale based on historical data, in order to predict the cluster of models for the new season and recommend actions to be carried out before the end of the season.
YEAR 2020-2021
CUSTOMER Fashion
METODOLOGIA
- Time Series Analysis
- ML models on the TimeSeries
- Forecasting
- Classification methods
- Statistical methods
Customer churn
Churn detection to identify and keep customers who intend to close their current account.
YEAR 2018
CUSTOMER Banks and Fintech
METODOLOGIA
- Forecast of basic current accounts closure through the identification of characteristic behaviours
- Grouping of movements by type
- Supervised Learning – Random Forest
Cryptocurrency trading
.Search for customer movements to/from online trading platforms for monitoring and proposing other investments
YEAR 2018
CUSTOMER Banks and FIntech
METODOLOGIA
- Identification of customers who made transactions to cryptocurrency exchange sites and analysis of transfers sent to/from those platforms
- Analysis of the correlation with the market prices of cryptocurrencies
- Application of Deep Learning algorithms to anticipate customers’ wishes (propensity to invest)
AI Toolchain
Definition of the hardware/ software infrastructure dedicated to development, testing and production of analytical models developed in the Bank
YEAR 2022-2025
CUSTOMER Banks and Fintech
METODOLOGIA
- Management of the entire life-cycle of an analytical model, starting from the development in a laboratory environment, passing through an industrialization environment up to the execution environment
- After release in production the model is monitored. When it no longer performs well it is trained again and the automated release process is repeated
Corporate Customer Analysis
Retail segmentation based on buying and selling trends to improve customer experience. Definition of a tool to predict the turnover that each retail could produce in the year-incoming season.
YEAR 2021
CUSTOMER Fashion
METODOLOGIA
- Data exploration of sell in and sell out data
- Clustering with basic techniques (k-means, DBSCAN) and advanced techniques (x-means, Gaussian Mixtures)
- Definition of similarity matrices and graph analysis
- Construction of a regressive model for estimating store purchase turnover, testing and validating multiple methods (linear regression, random forest, Elastic Net, SVR, GBRT)
Caronte
Immediate prioritisation of ticket processing to various IT teams.
YEAR 2019
CUSTOMER Fashion
METODOLOGIA
Adoption of a multi-step classifier through methodologies that consider words or word compositions as features, weighted according to their TF-IDF relevance
Aura Tutor for Fashion
System able to extract product information from various online catalogs of clothing brands and to output items similar to a text or visual query provided by the user.
YEAR 2022 – 2023
CUSTOMER Fashion
METODOLOGIA
- Dataset obtained from continuous scraping of text and images on websites in the fashion field
- Construction of an object retrieval model that makes extensive use of neural network CLIP (text and visual transformer)
Semantic Search Documents
Semantic search engine trained on banking. It is used for searching inside bank documents, providing in combination with a chatbot an internal staff support in solving problems.
YEAR 2018 – 2020
CUSTOMER Banks and Fintech
METODOLOGIA
- Dense network and word2vec algorithm
- Dataset extracted from public sources (Bankit, bancapedia, wikipedia, websites of major credit institutions)
Skill Alexa 7×24
Vocal use of the contents of Magazine 7×24 via Alexa.
YEAR 2019
CUSTOMER Publishing
METODOLOGIA
- Development of a custom skill to navigate the contents of an online newspaper through voice interaction (full or partial reading of the content)
- Implementation of the skill through the use of Nodejs and the framework provided by Amazon, both for logic and graphics
Sentiment Analysis
Sentiment Analysis on tweet for events. The system collects and analyzes tweets returning the sentiment towards the brand.
YEAR 2018
CUSTOMER Fashion
METODOLOGIA
- Developed using the Naive Bayes algorithm
Sales Forecasting
Sales forecasts for the e-commerce channel based on sales history and Google Analytics measurements.
YEAR 2019 – 2020
CUSTOMER Fashion
METODOLOGIA
- Dataset based on sales history and Google Analytics measurements to predict future sales and product appeal to recommend reorders (and avoid stock-outs) or reorganize shop windows to optimize sales
Warehouse of Parmesan cheese
.Tracking of Parmigiano cheese shapes in stock and their random repositioning due to an automatic cleaning process.
YEAR 2021 – 2022
CUSTOMER Food
METODOLOGIA
- OCR reading of the codes of interest (seasoning date, company registration number), correction of possible reading errors: systematic errors, heuristic kernel, similarities with reference templates
Information Extraction
Automatic classification of documents through images, information extraction.
YEAR 2017 – 2023
CUSTOMER Banks and Fintech
METODOLOGIA
- Deep Learning for image classification
- Feature template matching
- Graph-based document navigation
- Adaptation of Markov Chain to graphs
- OCR for text extraction
- Document classification for text documents
Press Review Analysis
Automatic Summary and Sentiment Analysis applied to articles included in press reviews.
YEAR 2017 – 2023
CUSTOMER Publishing
METODOLOGIA
- Polarity detection using BERT transformers neural network refinement
- Automatic summary based on text analytics extraction techniques