Georgia Loukatou

*TheFork, a TripAdvisor company, is a leading scaleup revolutionizing restaurant discovery and reservations throughout Europe. • Developing an intelligent multi-agent chatbot to enhance the user experience through context-aware, human-friendly recommendations. • Improving restaurant discovery by leveraging AI-powered recommendations enriched with user-generated content, ensuring personalised responses. • Working on LLM monitoring, observability, and testing for robust and reliable AI-driven interactions.

*Acolad is #1 European leader and among the top 10 leaders worldwide in language and content services. • Development of RAG,LLM and Prompt Engineering POCs to optimize document management and translation workflows. • Creation and deployment of an Agentic RAG solution for automatic post-translation editing, achieving a 40% improvement rate by automatically correcting poorly translated sentences without human intervention. • Design and implementation of an intelligent chatbot capable of leveraging the company’s internal documents to provide fast and accurate responses to employees, enhancing knowledge management. • Implementation of an LLM-as-a-Judge system for automatic and precise result validation. • Deployment of solutions from concept to production using MLOps best practices. • Collaboration with the Head of AI and the AI Product Manager to prioritize and define an AI roadmap.

* Sup de Vinci is a computer science school offering state-recognized training programs in engineering. Developed and taught a comprehensive 35-hour course on Natural Language Processing and Generative AI for Master 2 Data Science students, designing the curriculum from scratch, including course content, teaching materials, and hands-on practical sessions. Students explored transformers, LLMs, embeddings, prompt engineering, transfer learning and advanced textual analysis while working on real data such as product descriptions and reviews. They tackled challenges in noise handling, semantic understanding, sentiment extraction, and implemented Retrieval-Augmented AI Generation (RAG) and a conversational AI system, addressing both the ethical and technical aspects of GenAI. The program bridged NLP and AI concepts, highlighting use cases and respective pros and cons, while combining theoretical depth, implementation, and strategic brainstorming to prepare students for real-world applications and problem-solving.

*OpenClassrooms is an online education platform based in France providing computer science courses. As a mentor, successfully guided and supported several students in completing their projects, leading to the attainment of a Data Science certification.

*Qlip.ai is an AI-powered video highlights generator startup that helps repurpose long-form videos into shorter, shareable clips. Led Qlip.ai as its first hire, contributing to its growth within the vibrant startup ecosystem of Station F & Agoranov. Instrumental in securing initial funding and driving continuous client expansion. • Analyzed livestream and podcast recordings using advanced NLP and ML techniques, extracting valuable insights for decision-making. • Conducted robust data analysis and modeling, ensuring the preparation of high-quality multilingual datasets for modeling using SpaCy, Stanza, word and sentence embeddings. • Designed algorithms from scratch such as scoring and evaluating text chunks. • Developed models and NLP methods for information retrieval, NER, semantic and sentiment analysis, text classification, topic modeling, keywords, automatic title generation and summarization, text coherence. • Utilized transformers, open-source models, transfer learning, fine-tuning, HuggingFace. • Integrated LLMs (Mixtral 8x7b, Llama, Claude) for text comprehension tasks, using AWS (SageMaker, Comprehend, Bedrock) and Prompt Engineering. • Collaborated within a cross-functional team to address business use cases, and played a pivotal role in recruitment, development, and management of the research team. • Built experiments to increase model performance, incorporating client feedback for effectiveness. • Remained at the forefront of advancements in NLP Data Science, conducting cutting-edge research activities. • Excelled in a fast-paced startup environment, managing evolving priorities and meeting dynamic deadlines.

• Led a team of six in developing an open-source data processing pipeline focused on predicting the age of word learning, constructing a regression model based on word features. • Managed extensive datasets (1 billion words, 70,000 reports on lexical development, 17 languages). • Optimized a neural network model in Python to classify lexical categories in conversational speech, achieving an accuracy of over 85% on a hand-annotated test set in three languages. • Collaborated with the Stanford Human-Centered Artificial Intelligence group, implementing a lexical alignment model and measuring the complexity of player language in a reference game (MiC NeurIPS paper). • Presented findings in labs at Stanford University and University of California, Berkeley as invited speaker, as well as in international conferences.

• Initiated groundbreaking research on text-based word segmentation using unsupervised statistical models in collaboration with the University of Zurich, achieving up to a 30% increase in performance for eight low-resource languages. • Contributed to the development of a package containing text preprocessing tools, segmentation algorithms, and evaluation tools. • Led a team of four in collecting and organizing metadata for 320 public datasets. • Published several papers in international peer-reviewed conferences (ACL, CogSci, TALN) and three articles in high-quality journals (e.g., Cognition) as first author. • Trained 5 master students and 2 undergraduate students as interns.

Developed online grammars and dictionaries for diverse languages in the UNDL Foundation.

• Built a database of music scores from scratch at the General State Archives during fieldwork. • Participated in intensive seminars in American Culture and History, educational field trips, community service as well as other activities.