Freelance LISP : Nos derniers profils ayant déposé leur CV

Résumé d'expériences de Pierre,
freelance LISP habitant Paris (75)

• AI Expert

  Present (Paris, Freelance)
  Jan 2023 - aujourd'hui

  Working on a number of short-term freelance assignments mostly focused on Generative AI technologies. These
  include Retrieval-Augmented Generation (RAG), Supervised Fine Tuning (SFT) of both LLMs as well as Diffusion
  based models, and the application of Low-Rank Adaptation (LoRA) adapters to the FT process.

• Lead Data Scientist

  ADEO
  Jan 2022 - Jan 2023

  Managed a number of different ML projects related to recommender systems, product classification, product
  similarity and complementarity, knowledge graph enhancements and automatic completions using generative AI
  technologies. Specifically, I designed and implemented a Session-based Recommender System (SBRS) in
  TensorFlow based on several years of user purchase history. I designed, implemented and trained an advanced
  product classifier for 4,000+ product categories in the DIY industry. Implemented a Deep Learning based product
  similarity engine for the Knowledge Graph and the Recommendation Engine products. Designed a Graph Neural
  Network based recommender system based on a Graph Convolutional Network (GCN) architecture. Applied
  Generative AI approaches using the OpenAI API for the autocompletion of the company’s proprietary Knowledge
  graph and automatically linking Knowledge Graph concepts to client-facing website components. This work
  involved extensive use of the Cypher graph query language as well as Neo4j’s vast Graph Data Science (GDS)
  library. Finally, I was responsible for frequently advising product owners and business leaders on methodology
  and technology approaches and choices.

• Principal Data Scientist,

  CXUnified
  Jan 2020 - Jan 2022

  Responsible for all Machine Learning projects. Applied Deep Learning techniques to predict industry
  taxonomies for given web pages. This involved building a hierarchy of Deep Learning based multi-class
  classifiers in TensorFlow trained on millions of automatically generated and labeled training instances
  obtained by scraping Wikipedia and Amazon and then using semi-supervised learning techniques to generate
  additional training data from existing labeled training data. This resulted in millions of training instances. Also
  designed and developed unsupervised clustering algorithms to automatically establish sub-categories for
  retail industry taxonomies, specifically by using a modified version of hierarchical agglomerative clustering to
  not only infer sub-trees of the taxonomy but also automatically generate semantic labels for the nodes of the
  subtree. While some of this work involved designing proprietary algorithms, a significant portion involved
  using Deep Learning libraries such as Keras, Tensorflow, Gensim, NLTK and Scikit-Learn. All of this was
  done in Python in a Linux environment and deployed using Docker and Kubernetes.

• Principal Data Scientist

  Ayolab,
  Jan 2019 - Jan 2020

  Developed an AI based matching algorithm to classify, recognize and match web-based products to client’s
  master product list and identify gray market products. Part of this work involved product attribute extraction
  from product web pages. A tree of Deep Learning classifiers was also implemented using word embeddings
  and recurrent neural nets (LSTMs). This work involved knowledge modeling of ecommerce products, the
  development of a product taxonomy and the implementation of probabilistic classifiers for each node in the
  taxonomy as well as a multiphase product matching algorithm. Additional task involved managing and
  mentoring developers and data analysts.

• Director of Artificial Intelligence

  Jan 2017 - Jan 2018

  Managed a team of 5 individuals and in charge of all AI related projects. The primary project worked on
  involved detecting antisemitic hate speech of social media networks for the CRIF (Conseil Representative des
  Instituts Juives en France), a politically prominent organization representing 80+ Jewish Institutions in France.
  Designed and implemented a hate speech detection module originally using random forests. Bagging and
  boosting techniques were explored, but ultimately Deep Learning technologies were implemented and
  deployed. I was entirely responsible for the design, architecture. The implementation was divided amongst
  team members, with one member dedicated to the front end, two members dedicated to scraping various
  social media sites, and myself and a colleague worked all machine learning aspects as well as the overall
  system pipeline. This resulted in a comprehensive application that served to monitor and detect antisemitism
  on social networks in France. This employed a cascading neural net architecture using both Convolutional
  Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), specifically bi-directional LSTMs. Research
  aspects included experiments with training data augmentation approaches as well as automatic detection of
  new keywords based on word embeddings.

• Director of Machine Intelligence,

  Shareablee LLC
  Jan 2015 - Jan 2017

  Directed and managed the data science team and all machine learning projects. This includes, among other
  tasks, developing models for predicting if a Facebook post is promoted and models for classifying social
  media content by product type. This work was all done in Python, Pandas and Scikit-Learn
  Principal Architect and developer of the Audience Explorer product, which allows users to build custom
  audience segments consisting of millions of social media engagers. This particular project was completely
  implemented in Clojure & ClojureScript. This used Spark MLib to run distributed clustering algorithms on
  millions of data points consisting of social media users and their engagement metrics.
  Management responsibilities included writing up tasks and managing tasks, managing code reviews and
  training sessions, weekly one on ones, creating quarterly road maps and making hiring decisions with the
  CTO. Other major development tasks involved designing and implementing a distributed version of
  Hierarchical Agglomerative Clustering in Clojure for clustering users based social media engagement on
  various similarity metrics. Machine Learning Technologies included Python, Pandas, NLTK, Scikit-Learn,
  Spark ML, Hive, S3, GitHub, and Jira and Agile methodologies. Additionally several projects used Clojure and
  ClojureScript and associated libraries.
  Finally initiated and managed meetings and coordination with the Director of Client Success and her team in
  order to create, manage and prioritize projects and tasks and ensure a productive and pleasant interaction
  between my Data Science team and the Client Success team. Subsequently trained members of the Client
  Success team on running Hive queries and working in an agile environment using Jira to track and manage
  tasks.

• Senior Software Engineer,

  Jan 2013 - Jan 2015

  Designed and implemented new back-end components in Clojure. Designed and implemented a social media
  abstraction schema to facilitate knowledge abstraction and the generation of graph-based results for social
  media sites. Responsibilities also included managing and mentoring junior developers.
  Technology Manager, EyeCare Pro 2012 - 2013 (New York)
  • Responsibilities consisted of managing a team of six front and back-end developers as well maintaining and
  developing a Hunchentoot Common: Lisp webserver hosting approximately 1,000 client websites. This work
  involved working with Common Lisp, JavaScript, JQuery, Ajax and MySQL technologies and development
  tools.

• Chief Architect,

  Meperia LLC
  2009 - 2012

  Chief architect of BSG, an AI-based learning and understanding system for the assimilation of catalog data
  and knowledge in the context of the Healthcare Industry. Responsible for design decisions and prototypes.
  This involved extensive use Knowledge Representation, Machine Learning, Object-Oriented and functional
  programming design techniques. Reported to and work directly with the CTO.
  • Additional responsibilities involved designing, maintaining and enhancing the Prophet Quest system
  described below.

• Chief Architect

  Global Healthcare Exchange (GHX), Boulder, CO,
  2005 - 2009

  Principal architect of Prophet Quest, an AI Discernment System for the Health Care Industry. Extensive use
  Knowledge Representation and Machine Learning techniques in the context of a sophisticated object-oriented
  model. Responsible for all aspects of system design and prototyping, as well defining and writing up tasks for
  other team members. Also responsible for the generation of design documents and diagrams.

• Consultant,

  , Franz Inc., Oakland, CA
  2003 - 2004

  • Responsible for numerous activities including working with directly with prospective and existing clients,
  writing grant & contract proposals, attending trade shows, conferences and extensive networking. Principal
  organizer and Chairman of the International Lisp Conference 2002 (San Francisco) and the International Lisp
  Conference 2003 (New York). This work required extensive travel to France and Japan. Reported to and
  worked directly with the CEO.