My speech on "LLM-enhanced Graph in Private Equity" for Data innovation Summit, and more ...
At this year’s Data Innovation Summit (MEA), I presented on leveraging Knowledge Graphs (KGs) enhanced by Large Language Models (LLMs) to transform private equity decision-making. The talk highlighted EQT’s innovative use of proprietary and third-party data to construct dynamic KGs, enabling contextual retrieval and reasoning for deal sourcing, due diligence, and market analysis. I demonstrated how multimodal LLMs automate KG construction and improve retrieval processes, ensuring accurate and insightful responses to complex investment queries. The presentation concluded with practical applications and the potential of this technology to redefine AI-driven transformations in private equity. Please checkout the slide deck for more details.
Reflecting on my journey with Hyperight’s series of conferences, it has been an incredible opportunity to share insights, connect with industry leaders, and contribute to advancing knowledge in data science, machine learning, and AI-driven business innovation.
At the NDSML Summit in October 2023, I presented with Vilhelm von Ehrenheim on the topic “A Scalable and Adaptive System to Infer the Industry Sectors of Companies: Prompt + Model Tuning of Generative Language Models.” We introduced a novel system designed to infer industry sectors for companies by leveraging a hybrid approach of prompt and model tuning, optimized for generative language models like T5. This approach addressed critical challenges in private equity sector analysis, such as annotation scarcity, evolving sector frameworks, and high inference demands, showcasing its scalability and adaptability. The system demonstrated superior performance compared to traditional methods and has been effectively supporting hundreds of private equity professionals for over a year.
About 1 year further back at NDSML Summit 2022, Vilhelm von Ehrenheim and I presented our work, “Revenue Forecast for Growth Companies Using Scarce Time-Series Data,” addressing the challenges of limited financial data in high-growth companies. We introduced a scalable algorithm capable of long-term revenue extrapolation with confidence estimates, empowering investment professionals to make data-informed valuation and decision-making. The session also showcased how this algorithm was seamlessly integrated into an investment platform, providing practical insights into productionizing advanced analytics for private equity. Here is also an interview by Hyperight.
At the Data Innovation Summit 2022 in Stockholm, I co-presented with Sonja Horn on leveraging weakly-supervised NLP models to facilitate private capital operations. Our discussion centered on the PAUSE algorithm (Positive and Annealed Unlabeled Sentence Embedding), which was previously published at EMNLP 2021, and its integration into EQT’s proprietary Motherbrain platform. This innovative approach demonstrated how sentence embeddings, generated with minimal labeled data, could enhance tasks like deal sourcing and market analysis for various investment funds. The presentation highlighted the production deployment and practical applications of these models, illustrating their value in streamlining operations in private equity. A more technical presentation is available here.
Looking back, participating in Hyperight’s conferences like the Data Innovation Summit and NDSML Summit has been an incredibly rewarding journey. These events not only provided a platform to share my work but also created a space to connect with inspiring professionals and exchange groundbreaking ideas. The thoughtful curation of topics and the vibrant discussions make every Hyperight event a memorable and enriching experience. I truly appreciate the opportunity to be part of such a dynamic community driving data-driven innovation forward.