Develops energy-efficient large language models (LLMs) optimized for processing multimodal, long-context documents (32k+ tokens) in regulated industries. These models enable secure, local deployment on consumer GPUs and CPUs, ensuring compliance with EU legislation while providing explainable, auditable outputs for tasks like data harmonization and complex document analysis.
Funding
Funding not disclosed
Founders
Product
Problem
Organizations in regulated industries face challenges in processing large volumes of complex, multimodal documents while adhering to strict data privacy and compliance mandates, such as those imposed by EU legislation. Existing large language models (LLMs) often lack the energy efficiency, explainability, and local deployment options required to meet these stringent requirements.
Solution
Pleias develops energy-efficient LLMs optimized for processing multimodal, long-context documents (32k+ tokens) within regulated sectors. These models facilitate secure, local deployment on consumer-grade GPUs and CPUs, ensuring compliance with data protection laws. The AI assistants provide explainable and auditable outputs, citing the source documents used for each generation, which allows knowledge workers to verify the results. The models are designed for tasks such as data harmonization, complex document analysis, and Retrieval-Augmented Generation (RAG).
Target Audience
The primary target audience includes organizations in regulated industries, such as HR services and R&D, that require AI assistants for complex document processing, data harmonization, and analysis while adhering to strict compliance and data privacy standards.
Features
- Optimized for local deployment on consumer GPUs and CPUs, enabling offline mode for most tasks
- Designed for complex document processing, RAG, and data harmonization
- Explainable AI with precise citations to source documents for verification and reuse
- Compliant with EU legislation and suitable for sensitive use cases
- Excels at multilingual applications
- Models support long context windows (32k+ tokens)
- Custom classifier, Celadon, to detect toxic content in open data