Lexiq

Secure AI assistant for legal case analysis and discovery

Comment 1

1.⁠ Project Overview

  • Project Name: LexiQ
  • Purpose: LexiQ is a multi-agent AI-powered legal research platform aimed at providing comprehensive legal analysis through automation and AI. We were inspired by the inefficiency and privacy risks of modern legal research — most tools are either expensive or require sharing sensitive documents with AI models that don’t guarantee compliance. We wanted to build something better: a system that’s fast, interactive, and privacy-first. Through this project, we learned how to integrate multiple services — from Vanta’s masking validation to embedding generation and chat via Amazon Bedrock with Claude — into a secure, seamless pipeline. We also deepened our understanding of vector search using FAISS and how to enrich legal context with real-time news. How we build it :
  • AWS (Amazon Web Services):
    • S3: Storage for PDF files.
    • DynamoDB: Session and chat history management.
    • Bedrock: Hosting the Claude 3 Sonnet model for legal and conversational analysis.
  • Anthropic:
    • Claude 3 Sonnet: Utilized for advanced legal analysis and conversational interactions.
  • Vanta:
    • Compliance Logging: Automating compliance through PII redaction logs and audit trails.

4.⁠ ⁠Architecture Overview

  • Multi-Agent System: Specialized agents handle news relevance, statute reference, and judge bias analysis.
  • Vector Store: Uses FAISS for storing and retrieving case precedents.
  • Conversational AI: Chain-of-thought logic combined with RAG for enriched chat experiences.
  • Security: PII redaction, input validation, and hallucination detection ensure robust data protection and integrity. The biggest challenges we faced were reliably masking sensitive information before LLM interaction, grounding Claude’s responses in retrieved cases, and processing unstructured legal PDFs cleanly. Overcoming these helped us create LexiQ , a smarter, safer way to explore legal knowledge.

Built With

Share this project:

Updates