• 

  saved review

• 

  Interactive Findings

• 

  dashboard

• 

  new review

• 

  Interactive Findings

• 

  Interactive Findings

• 

  demo of decision review

• 

  interactive findings

DecisionGuard AI — Evidence-Based Decision Intelligence

The Problem

Organizations make high-stakes decisions using scattered policies, vendor proposals, contracts, and security documents. Important contradictions and missing requirements can go unnoticed when people review these documents manually. This can lead to security gaps, compliance risks, and poor decisions.

Our Solution

DecisionGuard AI transforms document review into an evidence-backed decision workflow. Users upload relevant documents, define a decision, and receive a structured analysis that highlights policy conflicts, missing requirements, security risks, and recommended remediation steps.

How It Works

1. Upload Evidence: Add PDF, DOCX, TXT, or Markdown documents containing policies, proposals, requirements, or assessments.
2. Retrieve Relevant Evidence: The RAG pipeline processes the documents and retrieves relevant passages to ground the analysis in source material.
3. Analyze with AI: NVIDIA Nemotron, accessed through Nebius Token Factory, compares the evidence across documents and generates structured findings.
4. Review Findings: Examine risk severity, an overall risk score, an executive summary, supporting evidence, and actionable recommendations.
5. Ask Follow-up Questions: Use the AI assistant to explore why a finding matters and which documents support it.
6. Export a Report: Generate a PDF decision review report for further discussion and documentation.

What Makes It Different

DecisionGuard AI is designed around the evidence behind a decision—not just a general-purpose chatbot response. It connects findings to source documents, highlights conflicts between requirements and proposals, and turns the analysis into a practical remediation checklist.

Technology

The application uses Python and Streamlit, a retrieval-augmented generation pipeline, document-processing libraries, local vector search, and NVIDIA Nemotron through Nebius Token Factory. SQLite supports saved reviews in the current implementation.

Demonstration

In our vendor-security scenario, DecisionGuard compares a company security policy with a vendor proposal. It identifies missing multi-factor authentication (MFA), a password-policy mismatch, and an incident-reporting timeframe conflict. The user can inspect the findings, ask follow-up questions, and download a PDF report.

Responsible Use

DecisionGuard AI is a decision-support tool. Its findings should be verified against the original evidence by qualified reviewers; it does not replace professional judgment or provide formal legal, compliance, or security certification.

Built With

• faiss
• hugging-face
• langchain
• nebius-token-factory
• nvidia-nemotron
• python
• retrieval-augmented-generation-(rag)
• sentence-transformers
• sqlite
• streamlit