Scribe is a data platform for company information, powered by research-based AI. Uncover deep insights about private companies and end data entry from PDFs. Welcome to the future of company knowledge.Scribe was born out of a desire to introduce cutting edge technology into finance, reducing administration and supporting business activity. Our mission is to explain the world's data.

Persuadr is the Quant of Customer Experience, helping financial services companies predict winning customer engagement strategies with the same rigor they use to pick winning assets. Our proprietary Persuasion Engine simulates all possible outcomes of a conversation, then recommends the optimal arguments and communication style to use with each customer. The Engine learns from every interaction, building rich and commercially valuable insights on customer profiles, needs and sentiment.

Broadly speaking, I am working on automatic persuasion methods for behaviour change.More precisely, we develop brand new method for: building the application domain, gathering data and opinion of the participants concerning the domain (nothing sneaky, everything is through surveys and discussions), learning a model of the participants, optimising the sequence of arguments played by our system during the discussion with the participants and updating the model of the participants though the discussion.For all of that we combine abstract argumentation, machine learning, decision theory, stats, and psychology.

My work consisted in developping size-reduction and valuation methods for the argumentation problems.

My research project was about mono/multi-agent sequential decision problems in non-stationary environments using Markov models.I applied traditional Markov decision models as well as my own on various domains, in particular the optimisation of strategic persuasive dialogues.

I worked on policy learning algorithms in non-stationary environment using regret minimization.

I worked on optimising the application of the Bucket Elimination method on inference algorithms for bayesian networks.