BEGIN:VCALENDAR VERSION:2.0 PRODID:-//datacraft - ECPv6.16.4//NONSGML v1.0//EN CALSCALE:GREGORIAN METHOD:PUBLISH X-WR-CALNAME:datacraft X-ORIGINAL-URL:https://datacraft.paris X-WR-CALDESC:Events for datacraft REFRESH-INTERVAL;VALUE=DURATION:PT1H X-Robots-Tag:noindex X-PUBLISHED-TTL:PT1H BEGIN:VTIMEZONE TZID:Europe/Paris BEGIN:DAYLIGHT TZOFFSETFROM:+0100 TZOFFSETTO:+0200 TZNAME:CEST DTSTART:20230326T010000 END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:+0200 TZOFFSETTO:+0100 TZNAME:CET DTSTART:20231029T010000 END:STANDARD BEGIN:DAYLIGHT TZOFFSETFROM:+0100 TZOFFSETTO:+0200 TZNAME:CEST DTSTART:20240331T010000 END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:+0200 TZOFFSETTO:+0100 TZNAME:CET DTSTART:20241027T010000 END:STANDARD BEGIN:DAYLIGHT TZOFFSETFROM:+0100 TZOFFSETTO:+0200 TZNAME:CEST DTSTART:20250330T010000 END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:+0200 TZOFFSETTO:+0100 TZNAME:CET DTSTART:20251026T010000 END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTART;TZID=Europe/Paris:20241210T140000 DTEND;TZID=Europe/Paris:20241210T180000 DTSTAMP:20260621T125332 CREATED:20240910T081234Z LAST-MODIFIED:20241204T141636Z UID:12756-1733839200-1733853600@datacraft.paris SUMMARY:Get insights from LLM chat data and monitor semantic drift DESCRIPTION:2406-Bunka-Bunka \n \n\n \n\n\n \n \n Inscription\n \n \n \nDescription \n\nIn the rapidly growing landscape of ChatGPT-like applications\, Retrieval-Augmented Generation (RAG) solutions are gaining significant traction. However\, analyzing the user-generated data from these applications to extract meaningful insights and adapt to emerging trends is a complex challenge. \nThis workshop introduces Bunka\, an open-source library designed for intuitive exploration and visualization of unstructured datasets. Using Bunka\, participants will explore embeddings in simplified projections\, empowering them to uncover hidden patterns and track semantic drift effectively. \nWe will work with a unique\, anonymized dataset of recently leaked chat channels from Andrew Tate’s private community\, tackling real-world challenges to extract actionable insights. The session will focus on techniques to:• Analyze user behavior and emerging trends in chat data.• Monitor semantic drift and its impact on application performance.• Adaptively predict and respond to new themes and trends. \nBy the end of this hands-on workshop\, you’ll gain practical experience with embedding-based analysis and actionable insights into leveraging semantic monitoring to enhance your LLM-driven applications. \n\nJoin us by registering now ! URL:https://datacraft.paris/event/exploring-filtering-datasets-for-llm-fine-tuning/ LOCATION:datacraft –\, 3 rue Rossini\, 75009 Paris\, France CATEGORIES:Hybrid event,in English ORGANIZER;CN="datacraft":MAILTO:contact@datacraft.paris END:VEVENT BEGIN:VEVENT DTSTART;TZID=Europe/Paris:20241204T090000 DTEND;TZID=Europe/Paris:20241204T180000 DTSTAMP:20260621T125332 CREATED:20241008T132124Z LAST-MODIFIED:20241104T143320Z UID:12994-1733302800-1733335200@datacraft.paris SUMMARY:Unlocking Vector Databases : Search and RAG with Vespa DESCRIPTION:Inscription\n \n \n \n2412-Vespa-VectorSearch \nIn this workshop\, we will dive in Vector Databases using Vespa\, an open-source search engine and vector/tensor DB used in various major projects such as Spotify or Vinted recommandation system. The objective is to give participants tools to work on various use-cases form E-commerce search to recommendations and RAG. \nThe workshop is made of 6 main chapters : \n\nIntroduction: 10km view of the use-cases and anti-use-cases for Vespa and why. Lab will be centered around CRUD\nLexical Search: query syntax\, boolean clauses\, ranking for product search\nGrouping: facets for data exploration on sales data and result diversification\nTensors: use tensor math for recommendations and vector search on an E-commerce dataset\nHybrid Search: generate embeddings on the fly; use RRF; implement RAG\nPerformance and scaling: intro to how Vespa scales; typical performance bottlenecks\n\nEach chapter is composed of : \n\n a short introduction to the topic\na lab that will be solved in small groups\, working on a pre-defined dataset\na debriefing with the large group\nshort wrap-up\n\nWarning : Even though online attendance is opened for this event\, we highly encourage participants to attend on site to get the most out of this event (participants will form groups and work together during the workshop). URL:https://datacraft.paris/event/unlocking-vector-databases-search-and-rag-with-vespa/ LOCATION:datacraft –\, 3 rue Rossini\, 75009 Paris\, France CATEGORIES:Hybrid event,in English END:VEVENT BEGIN:VEVENT DTSTART;TZID=Europe/Paris:20241115T150000 DTEND;TZID=Europe/Paris:20241115T170000 DTSTAMP:20260621T125332 CREATED:20241016T152153Z LAST-MODIFIED:20241016T170520Z UID:13195-1731682800-1731690000@datacraft.paris SUMMARY:Responsible Prompting – Real-Time Prompt Recommendation DESCRIPTION:2411-IBM-ResponsiblePrompting\n \n\n \n\n\n \n \n Inscription\n \n \n \n\nJoin us for a demo and a hands-on exercise with this open-source tool called Responsible Prompting\, developed by the Responsible Tech Team\, IBM Research. This tool supports your LLM users in crafting more responsible prompts by recommending good practices and preventing harmful prompts by leveraging the semantics mapping provided by sentence transformers. The recommendations are provided in prompting-time\, i.e.\, before users send the prompt to an LLM. Hence\, for many tasks\, it has the potential to speed up prompting tasks\, improve the quality of prompts\, and even reduce costs. The tool was designed to be easy to customize to different business cases. And it’s even possible go even further by fine-tuning the backbone sentence transformers models.We will go through the following steps:\n– Quick demo of the tool (10 min)\n– Deep dive into the project history and underlying model (sentence transformers) (20 min)\n– Discussion on how it can help businesses (esp. big companies) support their users and diminish the misuses of these technologies. (10 min)\n– Conclusion\, question and discussion with the attendees (10-20 min)(optional) Bonus: a hands-on exercise for customizing the tool (60 minutes).\nThis event will be in English and in hybrid format. URL:https://datacraft.paris/event/responsible-prompting-real-time-prompt-recommendation/ LOCATION:datacraft –\, 3 rue Rossini\, 75009 Paris\, France CATEGORIES:Hybrid event,in English END:VEVENT BEGIN:VEVENT DTSTART;TZID=Europe/Paris:20241112T180000 DTEND;TZID=Europe/Paris:20241112T190000 DTSTAMP:20260621T125332 CREATED:20240923T101432Z LAST-MODIFIED:20241001T065047Z UID:12881-1731434400-1731438000@datacraft.paris SUMMARY:REX: Amorcer la transformation GenAI en scale-up DESCRIPTION:Inscription\n \n \n \n2411-Tremblay-GenAIScaleUp \nDescription\nDans un contexte où les scale-ups sont passées d’un objectif d’hypercroissance à un objectif de rentabilité\, comment assurer leur transformation GenAI de façon optimale ?\nCe talk offre un premier retour d’expérience 6 mois après le lancement de ce chantier chez Ornikar par l’équipe AI. On fait le point sur les choix technologiques\, l’organisation mise en place\, les use-cases explorés ainsi que les défis relevés en chemin. Quelques enseignements à partager et des questions en suspens pour échanger collectivement.\nÀ propos du speaker Charles Tremblay :\nData scientist et actuaire\, Charles est Head of Artificial Intelligence chez Ornikar\, où il travaille sur des sujets d’InsurTech\, de connaissance client et de Generative AI. URL:https://datacraft.paris/event/rex-amorcer-la-transformation-genai-en-scale-up/ LOCATION:datacraft –\, 3 rue Rossini\, 75009 Paris\, France CATEGORIES:Hybrid event,in French ORGANIZER;CN="datacraft":MAILTO:contact@datacraft.paris END:VEVENT BEGIN:VEVENT DTSTART;TZID=Europe/Paris:20241016T183000 DTEND;TZID=Europe/Paris:20241016T193000 DTSTAMP:20260621T125332 CREATED:20240830T164110Z LAST-MODIFIED:20241205T132545Z UID:12665-1729103400-1729107000@datacraft.paris SUMMARY:Monitoring et Évaluation des systèmes LLM DESCRIPTION:2410-Saxce-EvalLLM \n \n\n \n\n\n \n \n Inscription\n \n \n \nIntervenant  \nHugues de Saxcé\, AI Engineer chez Chainlit by Literal AI \nDescription \nL’évolution rapide des modèles de langage (LLM) pose des défis importants pour les développeurs en termes de monitoring\, d’observabilité et d’évaluation continue. Cet atelier abordera ces problématiques la présentation un framework innovant permettant aux développeurs de surveiller efficacement les performances de leurs applications LLM\, en particulier dans le contexte de systèmes conversationnels. \nChainlit a mis au point un SDKs en Python ou Typescript pour loguer chaque étape du flux d’une application\, offrant ainsi une visibilité complète sur des aspects clés comme la gestion des tokens\, le timing\, et la détection d’erreurs. \nUne attention particulière est portée sur la partie évaluation\, essentielle pour prévenir les baisses de performance lors des modifications de modèles ou d’autres composants d’application. \nHugues partagera également des insights sur les meilleures pratiques pour construire et exploiter des datasets d’évaluation\, garantissant que les ajustements apportés ne compromettent pas la qualité du système. \nCible \nCet atelier est idéal pour les data scientist\, AI/ML engineer et les développeurs souhaitant maîtriser les outils et techniques de suivi et d’évaluation des LLM pour des déploiements robustes et fiables. \n  URL:https://datacraft.paris/event/monitoring-et-evaluation-des-systemes-llm/ LOCATION:datacraft –\, 3 rue Rossini\, 75009 Paris\, France CATEGORIES:Hybrid event,in French ORGANIZER;CN="datacraft":MAILTO:contact@datacraft.paris END:VEVENT END:VCALENDAR