PyData Seattle 2023

You Want to Buy This - Particle Swarm Classification for Next-Gen Recommendation Engines
04-28, 11:00–11:45 (America/Los_Angeles), Hood

Case study that describes how a scrappy science and engineering team built an optimal recommendations engine for consumer banking and FinTech mobile app users. The engine produces high-response, tailored end-user results from anonymized and incomplete data, the application of quantum particle swarm optimization techniques, and by leveraging a homegrown knowledge representation graph.

Current banking apps frustrate consumers by recommending useless products and services that don’t fit the user preferences, lifestyle, or spending patterns. Next-Gen recommendation engines give results that lead to optimal payment card use, raise consumer satisfaction and cash-back totals, improve banking relationships, and lead to higher revenue for banks and merchants (mom-n-pop to big box stores). This case study describes how a scrappy science and engineering team built the CLOE recommendation engine. CLOE produces high-response, tailored end-user results from models built with highly anonymized and incomplete payment data, applies quantum particle swarm optimization techniques to identify localized search spaces, and leverages a homegrown and standards-based KDR graph.

24 slides, about 1 minute per slide.


• Use case and background
• The team – who, what, how
• The technology
• Why quantum particle swarm optimization?
• Fully informed PSOs
• Probability distributions and the real world
• What is a Quantum PSO? Comparison vs PSO
• Advantages: better results, easier implementation
• Clustering and classification
• Diversity control == no premature convergence
• Behavior analysis and convergence
• Implementation, team, and tech
• Before and after: results
• The future: small world dynamics, neural networks training
• Q&A

Particle swarm optimization gets less attention than other classification and clustering techniques, yet it can be applied through cheaper and faster computational resources, and with simpler data models, than using ML techniques in vogue. End results are equivalent but at a fraction of the implementation and training time, and with reproducible results.

Prior Knowledge Expected

No previous knowledge expected

Eugene Ciurana is the CTO of Triple (, the leading provider of next-gen CLO technology to the largest US and European banks, and to some of the largest content and ad networks in the US. Prior to Triple, Eugene was the senior director of knowledge discovery and representation at Meltwater US1, managing science and engineering teams in San Francisco, London, Stockholm, and Budapest; he was CEO and founder of Cosmify, Inc. “one of the last pure AI company acquisitions in Silicon Valley” in 2017 and known as “Palantir in a box.” Before that he was the CTO of Summly, the most successful automated text summarization company in Silicon Valley history, Sr VP of technology at Badoo/Bumble, director of systems integration at LeapFrog Enterprises, and chief architect at Global. Eugene can be reached on the Libera and OFTC IRC networks (#vim, #python, #java, #awk, #wikimedia, #tor) under the /nick pr3d4t0r. Twitter: @ciurana