Data Science Conversations Damien Deighan and Philipp Diesinger

In this episode we talk to Kirk Marple about the power of Knowledge Graphs when combined with GenAI models.  Kirk explained the growing relevance of knowledge graphs in the AI era, the practical applications, their integration with LLMs, and the future potential of Graph RAG.
Kirk Marple a veteran of Microsoft and General Motors, Kirk has spent the last 30 years in software development and data leadership roles. He also successfully exited the first startup he founded, RadiantGrid, acquired by Wohler Technologies.
Now, as the technical founder and CEO of Graphlit, Kirk and his team are streamlining the development of vertical AI apps with their end-to-end, cloud based offering that ingests unstructured data and leverages retrieval augmented generation to improve accuracy, domain specificity, adaptability, and context understanding – all while expediting development.
Episode Summary -
Introduction to Knowledge Graphs:Knowledge graphs extract relationships between entities like people, places, and things, facilitating efficient information retrieval.They represent intricate interactions and interrelationships, enabling users to "walk the graph" and uncover deeper insights.Importance in the AI Era:Knowledge graphs enhance data retrieval and filtering, crucial for feeding accurate data into large language models (LLMs) and multimodal models.They provide an additional axis for information retrieval, complementing vector search.Industry Use Cases:Commonly used in customer data platforms and CRM models to map relationships within and between companies.Knowledge graphs can convert complex datasets into structured, easily queryable formats.Challenges and Limitations:Familiarity with graph databases and the ETL process for graph data integration is still developing.Graph structures are less common and more complex than traditional relational models.Integrating Knowledge Graphs with LLMs:Knowledge graphs enrich data integration and semantic understanding, adding context to text retrieved by LLMs.They can help reduce hallucinations in LLMs by grounding responses with more accurate and comprehensive context.Graph RAG (Retrieval Augmented Generation):Combines knowledge graphs with RAG to provide additional context for LLM-generated responses.Allows retrieval of data not directly cited in the text, enhancing the breadth of information available for queries.Scalability and Efficiency:Effective graph database architectures can handle large-scale graph data efficiently.Graph RAG requires a robust ingestion pipeline and careful management of data freshness and retrieval processes.Future Developments:Growing interest and implementation of knowledge graphs and Graph RAG in various industries.Potential for new tools and standardization efforts to make these technologies more accessible and effective.Graphlit: Simplifying Knowledge Graphs:The platform focuses on simplifying the creation and use of knowledge graphs for developers.Provides APIs for easy integration, supporting domain-specific vertical AI applications.Offers a unified pipeline for data ingestion, extraction, and knowledge graph construction.Open Source and Community Contributions:Recommendations for...

At LanguageTool, Bartmoss St Clair (Head of AI) is pioneering the use of Large Language Models (LLMs) for grammatical error correction (GEC), moving away from the tool's initial non-AI approach to create a system capable of catching and correcting errors across multiple languages.
LanguageTool supports over 30 languages, has several million users, and over 4 million installations of its browser add-on, benefiting from a diverse team of employees from around the world.
Episode Summary -
LanguageTool decided against using existing LLMs like GPT-3 or GPT-4 due to cost, speed, and accuracy benefits of developing their own models, focusing on creating a balance between performance, speed, and cost.The tool is designed to work with low latency for real-time applications, catering to a wide range of users including academics and businesses, with the aim to balance accurate grammar correction without being intrusive.Bartmoss discussed the nuanced approach to grammar correction, acknowledging that language evolves and user preferences may vary, necessitating a balance between strict grammatical rules and user acceptability.The company employs a mix of decoder and encoder-decoder models depending on the task, with a focus on contextual understanding and the challenges of maintaining the original meaning of text while correcting grammar.A hybrid system that combines rule-based algorithms with machine learning is used to provide nuanced grammar corrections and explanations for the corrections, enhancing user understanding and trust.LanguageTool is developing a generalized GEC system, incorporating legacy rules and machine learning for comprehensive error correction across various types of text.Training models involve a mix of user data, expert-annotated data, and synthetic data, aiming to reflect real user error patterns for effective correction.The company has built tools to benchmark GEC tasks, focusing on precision, recall, and user feedback to guide quality improvements.Introduction of LLMs has expanded LanguageTool's capabilities, including rewriting and rephrasing, and improved error detection beyond simple grammatical rules.Despite the higher costs associated with LLMs and hosting infrastructure, the investment is seen as worthwhile for improving user experience and conversion rates for premium products.Bartmoss speculates on the future impact of LLMs on language evolution, noting their current influence and the importance of adapting to changes in language use over time.LanguageTool prioritizes privacy and data security, avoiding external APIs for grammatical error correction and developing their systems in-house with open-source models.

In this enlightening episode, Dr. Julia Stoyanovich delves into the world of responsible AI, exploring the ethical, societal, and technological implications of AI systems. She underscores the importance of global regulations, human-centric decision-making, and the proactive management of biases and risks associated with AI deployment. Through her expert lens, Dr. Stoyanovich advocates for a future where AI is not only innovative but also equitable, transparent, and aligned with human values.
Julia is an Institute Associate Professor at NYU in both the Tandon School of Engineering, and the Center for Data Science.  In addition she is Director of the Center for Responsible AI also at NYU.  Her research focuses on responsible data management, fairness, diversity, transparency, and data protection in all stages of the data science lifecycle. 
Episode Summary -
The Definition of Responsible AIExample of ethical AI in the medical world - Fast MRI technologyFairness and Diversity in AIThe role of regulation - What it can and can’t doTransparency, Bias in AI models and Data ProtectionThe dangers of Gen AI Hype and problematic AI narratives from the tech industryThe impotence of humans in ensuring ethical development Why “Responsible AI” is actually a bit of a misleading termWhat Data & AI leaders can do to practise Responsible AI

Luis Moreira-Matias is Senior Director of Artificial Intelligence at sennder, Europe’s leading digital freight forwarder. At sennder, Luis founded sennAI: sennder’s organization that oversees the creation (from R&D to real-world productization) of proprietary AI technology for the road logistics industry.

During his 15 years of career, Luis led 50+ FTEs across 4+ organisations to develop award-winning ML solutions to address real-world problems in various fields such as e-commerce, travel, logistics, and finance. 

Luis holds a Ph.D. in Machine Learning from the U. Porto, Portugal. He possesses a world-class academic track with high impact publications at top tier venues in ML/AI fundamentals, 5 patents and multiple keynotes worldwide - ranging from Brisbane (Australia) to Las Palmas (Spain).

In this episode Tarush Aggarwal, formerly of Salesforce and WeWork is back on the podcast to discuss the evolution of the Semantic layer and how that can help practitioners get results from LLMs.  We also discuss how smaller ELMs (expert language models) might be the future when it comes to consistent reliable outputs from Generative AI and also the impact of all of this on traditional BI tools.

In this episode Patrick McQuillan shares his innovative Biological Model - a concept you can use to enhance data outcome in large enterprises.  The concept takes the idea that the best way to design a data strategy is to align it closely with a biological system.
He discusses the power of centralized information, importance of data governance, and the necessity for a common performance narrative across an organization.
Episode Summary -
- Biological Model Concept
- Centralized vs. Decentralized Data
- Data Collection and Maturity
- Horizontal translation layer 
- Partnership with vertical leaders
 - Curated data layers 
- Data dictionary for consistency
- Focusing on vital metrics
- Data Flow in Organizations
- Biological Model Governance
- Overcoming Inconsistency and Inaccuracy