Leading the technology development at Griot.

We deliver bespoke AI systems that help companies extract more value from their data, working as a freelance consultancy or software development service. We provide advice on innovation strategies related to AI adoption for a variety of industries.

I was part of the NLP team for the Question Answering system developed at Planting Space. Amongst other teams we developed the training pipeline for LLMs to translate the query in natural language into a DSL that our symbolic system could reason about.

I've lead the data science effort of company after multiple acquisitions. In addition to NLP and ML based products built so far, we started maintaining and extending products related to data experimentation and analysis. Our team maintained and extended existing data pipelines for A/B testing, acceleration of statistical significance, experimentation, and recommendation. We unified a variety of cloud data sources through the usage of Docker, Kubernetes, Terraform, and Luigi.

After Episerver's acquisition of Idio, the data science teams of both companies were unified under my management. We are responsible for maintain and grow the data science effort of the joint company in the areas of Machine Learning, Natural Language Processing and Recommendation Systems through robust data pipelines, using luigi, spark, python and scala as technolgies.

As VP of Research and Development, I'm responsible for leading a team of research engineers that maintain and grow Idio's NLP pipeline and recommendation engine. As a team, we draw from state of the art research to implement new features for Idio's AI ecosystem. We are responsible for taking a proof-of-concept to production and we often read/write papers as part of the process.

I maintained Idio's Natural Language Processing pipeline and recommendation engine. The codebase is mostly based in Scala and Python and the ecosystem relied heavily on batch processing via Spark and Luigi. These batch processing tools also were aimed at maintaining a Knowledge Graph (stored in Neo4J) besides models for Entity Linking. We often drew from state of the art research for extending or adding new features to Idio's ecosystem.

I give advice and provide mentorship on Runway's NLP pipelines. Runway provides a platform for safe messaging between a company's employees and directors.

I've tutored on a project related to the generation of textual embeddings based on a short description about an entity and how that can improve the generation of OOV vectorial representations for a variety of downstream NLP tasks.

I have contributed to the development of the open source entity linking system DBpedia Spotlight and related libraries.

I took part of a DBpedia Association project that aimed to introduce dense entity embeddings and a log linear disambiguation model for the entity linking system DBpedia Spotlight.

I was responsible for developing a prototype that aimed to assist fact checking by linking claims in new articles to information available in the world bank and other public sources.

ComplyAdvantage uses Natural Language Processing and Machine Learning to identify individuals that might be politically exposed or involved in some criminal activity. I helped design and develop a set of machine learning tools capable of inferring and extracting that information from the web. The resulting (big) data set is served through Comply Advantage's API to help other institutions to fight fraud and prevent the funding of illicit activities.

I was part of a team that developed a smart assistant for mobile devices capable of recommending content (papers, books, movies, exhibitions) as well as keep track of educational activities for university students in order to improve their learning experience.

I was part of a data science team that developed and maintained the technology responsible for matching news articles with user's reading preferences. We also maintained data pipelines that enriched news articles with metadata as part of Signal's news tracking service. Most of the infrastructure was written in Clojure.

Idio's R&D team was responsible for developing a NLP engine for extracting metadata to be used in a semantic recommendation engine / personalisation platform (within the content marketing / marketing automation industry). In addition to these components, we also maintained a Knowledge Graph to infer extra information from textual content. We often had to draw from state of the art research to develop extra features for these components.

I was part of a team that aims to develop cloud-based solutions to the public sector, using standard web-development tools - javascript, html, etc - as well as platform-as-a-service tools (Salesforce).

I developed algorithms that extract people's opinions about a product or service from raw text in large volumes of data (social networks, forums, (micro) blogs, etc...). I worked on the implementation of a grammar of sentiment terms of Portuguese that became part of IBM's Cognos Consumer Insight (CGI).

I taught tutorials for BAs (first year linguistics students) and BScs (second yearpsychology students) on introduction to Psycholinguistics. I also took care of the module’ s online activities via Moodle. The course was about the problem faced by the child as a learner of a first language.

I lectured on informal logic, introductory ethics, and epistemology to BA students of a variety of degrees. One of the aims of Pitágoras University (Faculdade Pitágoras) was to give a background in humanities regardless of the degree taken.

I assisted Dr. Paulo Roberto Margutti Pinto in the co-ordination of two research groups at UFMG: logic lab (see undergraduate research position below for a description) and Brazilian Philosophy. The job involved the selection of readings and organisation of meetings. The Brazilian Philosophy group aimed to extract and systematize philosophical positions from Brazilian intellectuals from the colonial period onwards.

I participated in a project called 'Logic Lab'. It aimed to create computer assisted teaching methods for the logic modules of the BA Philosophy degree and a webpage about the software was also created at the time. The team used Visual Basic to implement software that checked the validity of proofs automatically and HTLM to provide a download base and related documentation.

I installed software and hardware from internet providers that allowed home users to connect to the internet via a dial-up network.