Samu Toljamo

LLM Aggregators

LLM Aggregators: Samu Toljamo, Olli Glorioso, Viljami Hakkarainen and David Ramos\n\nOur Three-Step Approach:\n\nStep 1: Group Similar Innovations using embeddings\n\nGenerate semantic embeddings from innovation descriptions and titles\nUse similarity thresholds to identify potential duplicate clusters\nScale analysis across thousands of innovation records\n\nStep 2: Validate Groups with LLM\n\nAzure OpenAI reviews each cluster for false positives\nRemoves incorrectly grouped innovations with detailed reasoning\nEnsures high precision while maintaining recall\nStep 3: Aggregate Results with LLM\n\nLLM combines information from multiple sources about the same innovation\nCreates unified innovation profiles preserving all source details\nMaintains full traceability while consolidating descriptions\n\nYou will find some visualizations from the project link and in the repo the most interesting file is the main.ipynb file.\n\nVideo: https://drive.google.com/drive/folders/1ZdlPXga2n17u7B9Z9KeLhf-8IOcioF5p?usp=sharing