Skip to main content

Advaizr AI

Advaizr leverages state-of-the-art AI models carefully selected for specific tasks within legal and financial workflows.

Core AI Technologies​

Advaizr employs a strategic combination of specialized AI models:

LLM Selection​

  • Gemini 1.5 Pro: Used for retrieval and high-context tasks, leveraging its expansive context window for comprehensive analysis of large documents and datasets.

  • DeepSeek-V2 (Reasoning Variant): Deployed for complex analysis tasks that require deep logical reasoning, such as auditing, contract comparison, and financial modeling.

  • DeepSeek-V2 (General Purpose/Chat): Utilized for drafting, summarization, and general interaction, providing consistent and high-quality outputs for standard tasks.

Vector Embeddings​

  • BGE-M3: Our embedding model of choice, selected for its state-of-the-art performance and hybrid search capabilities (dense and sparse vectors).

Backend Systems​

  • Milvus Vector Database: Configured for hybrid search to fully leverage BGE-M3's capabilities, enabling precise retrieval of relevant information.

  • LangGraph: Powers our multi-agent architecture, orchestrating complex workflows between specialized agents.

  • CopilotKit: Enables our dynamic, generative UI that adapts to user needs with interactive components.

Model Selection Philosophy​

Our AI stack is designed with several key principles:

  1. Task Specialization: Different models are deployed based on the specific requirements of each task.

  2. Context Optimization: Models with larger context windows handle document-heavy tasks.

  3. Reasoning Depth: Complex analytical tasks are assigned to models with superior reasoning capabilities.

  4. Output Quality: Generation tasks are routed to models that excel in producing clear, coherent content.

Continuous Improvement​

Advaizr's AI capabilities are regularly evaluated and updated:

  • Benchmarking against industry standards
  • Integrating new models as they become available
  • Fine-tuning for domain-specific performance
  • Collecting user feedback to improve accuracy

Learn more about how these AI technologies drive our core platform components in the Nexus and Hub documentation.