Sluicebox developed proprietary artificial intelligence agents to build a structured model for products and supply chains, tackling complex data pipeline problems.
Dynamic LCA Data Ingestion
The system begins by ingesting multimodal records and converting them into coherent product data. It enriches incomplete material lists with information such as component weight, material composition, and manufacturing activity. This gives teams a clear view of product contents, emission origins, and data gaps requiring supplier input.
The company automated three labor-intensive phases into AI workflows. Component classification sorts each bill of materials row into an approximately 100-node taxonomy. Partial supplier data extraction turns unstructured emails and documents into clean records. Emission factor matching maps each component or process to the best available factor from an inventory containing over 10,000 entries.
Deploying Lucy for Supplier Data
The team also built an AI supplier agent named Lucy to collect and validate primary data. Lucy guides suppliers through information requests, identifies missing fields, and checks if values fall outside expected ranges.
This approach replaces industry averages with data based on how components were actually manufactured. Companies can achieve primary data coverage from over 90% of their suppliers in a matter of weeks rather than years.
Integrating NVIDIA Nemotron 3 Ultra
Sluicebox considered NVIDIA Nemotron 3 Ultra for these workflows. NVIDIA Nemotron is a family of open models designed for efficient agentic reasoning and rapid task completion.
In Sluicebox tests, Nemotron 3 Ultra processed the context needed to classify parts, structure supplier data, and select emission factors. It delivered an inference cost reduction of 51% to 80%, improved accuracy, and response times equal to or faster than the company’s previous production model.
Maintaining Context in Dynamic Conversations
Data collection functions as an ongoing conversation rather than a one-off calculation. Delayed responses can interrupt information exchange and complicate an already challenging process.
Faster inference allows Lucy to absorb new information, maintain context, flag potential outliers, and guide suppliers toward complete responses while interactions remain active.
