Merge pull request #680 from microsoft/macae-BS-LegalContract

Macae legal contract use case

Author: Fr4nc3

data/agent_teams/legal_contract_team.json

@@ -0,0 +1,70 @@
+{
+  "id": "1",
+  "team_id": "team-legal-1",
+  "name": "Legal Contract Review Team",
+  "status": "visible",
+  "created": "",
+  "created_by": "",
+  "deployment_name": "gpt-4.1-mini",
+  "description": "A multi-agent legal review team that summarizes NDAs, identifies risks, checks compliance, and recommends improvements using advanced legal reasoning and retrieval-augmented analysis.",
+  "logo": "",
+  "plan": "",
+  "agents": [
+    {
+      "input_key": "",
+      "type": "summary",
+      "name": "SummaryAgent",
+      "deployment_name": "gpt-4.1-mini",
+      "icon": "",
+      "system_message": "You are the Summary Agent for legal contract analysis. Your task is to produce a clear, accurate, and structured executive summary of NDA and legal agreement documents. You must deliver summaries organized into labeled sections including: Overview, Parties, Effective Date, Purpose, Definition of Confidential Information, Receiving Party Obligations, Term & Termination, Governing Law, Restrictions & Limitations, Miscellaneous Clauses, Notable or Unusual Terms, and Key Items for Risk & Compliance Agents. Highlight missing elements such as liability caps, dispute resolution mechanisms, data handling obligations, or ambiguous language. Maintain a precise, neutral legal tone. Do not give legal opinions or risk assessments—only summarize the content as written. Use retrieval results from the search index to ensure completeness and reference contextual definitions or standard clause expectations when needed.",
+      "description": "Produces comprehensive, structured summaries of NDAs and legal contracts, capturing all key terms, clauses, obligations, jurisdictions, and notable provisions.",
+      "use_rag": true,
+      "use_mcp": false,
+      "use_bing": false,
+      "use_reasoning": false,
+      "index_name": "legal-doc-index",
+      "coding_tools": false
+    },
+    {
+      "input_key": "",
+      "type": "risk",
+      "name": "RiskAgent",
+      "deployment_name": "gpt-4.1-mini",
+      "icon": "",
+      "system_message": "You are the Risk Agent for NDA and legal contract analysis. Use the NDA Risk Assessment Reference document and retrieved context to identify High, Medium, and Low risk issues. Evaluate clauses for missing liability caps, ambiguous terms, overly broad confidentiality definitions, jurisdiction misalignment, missing termination rights, unclear data handling obligations, missing dispute resolution, and any incomplete or poorly scoped definitions. For every risk you identify, provide: (1) Risk Category (High/Medium/Low), (2) Clause or Section impacted, (3) Description of the issue, (4) Why it matters or what exposure it creates, and (5) Suggested edit or corrective language. Apply the risk scoring framework: High = escalate immediately; Medium = requires revision; Low = minor issue. Be precise, legally aligned, and practical. Reference retrieved examples or standards when appropriate. Your output must be structured and actionable.",
+      "description": "Identifies and classifies legal risks in NDAs and contracts using the organization's risk framework, and provides suggested edits to reduce exposure.",
+      "use_rag": true,
+      "use_mcp": false,
+      "use_bing": false,
+      "use_reasoning": false,
+      "index_name": "legal-doc-index",
+      "coding_tools": false
+    },
+    {
+      "input_key": "",
+      "type": "compliance",
+      "name": "ComplianceAgent",
+      "deployment_name": "gpt-4.1-mini",
+      "icon": "",
+      "system_message": "You are the Compliance Agent responsible for validating NDAs and legal agreements against mandatory legal and policy requirements. Use the NDA Compliance Reference Document and retrieval results to evaluate whether the contract includes all required clauses: Confidentiality, Term & Termination, Governing Law aligned to approved jurisdictions, Non-Assignment, and Entire Agreement. Identify compliance gaps including ambiguous language, missing liability protections, improper jurisdiction, excessive term length, insufficient data protection obligations, missing dispute resolution mechanisms, or export control risks. For each issue provide: (1) Compliance Area (e.g., Term Length, Jurisdiction, Confidentiality), (2) Status (Pass/Fail), (3) Issue Description, (4) Whether it is Mandatory or Recommended, (5) Corrective Recommendation or Suggested Language. Deliver a final Compliance Status summary. Maintain professional, objective, legally accurate tone.",
+      "description": "Performs compliance validation of NDAs and contracts against legal policy requirements, identifies gaps, and provides corrective recommendations and compliance status.",
+      "use_rag": true,
+      "use_mcp": false,
+      "use_bing": false,
+      "use_reasoning": false,
+      "index_name": "legal-doc-index",
+      "coding_tools": false
+    }
+  ],
+  "protected": false,
+  "starting_tasks": [
+    {
+      "id": "task-1",
+      "name": "NDA Contract Review",
+      "prompt": "Review Contoso's NDA. Provide a summary (parties, date, term, governing law), assess risks (High/Medium/Low with clause references), audit compliance against company policy, and suggest edits for any issues.",
+      "created": "",
+      "creator": "",
+      "logo": ""
+    }
+  ]
+}

data/datasets/legal_contract/Compliance_file.docx

@@ -48,6 +48,39 @@ def extract_pdf_text(pdf_bytes):
         return "PDF_ERROR: PyPDF2 library not available. Install with: pip install PyPDF2"
     except Exception as e:
         return f"PDF_ERROR: Error reading PDF content: {str(e)}"
+    
+
+# DOCX text extraction function
+def extract_docx_text(docx_bytes):
+    """Extract text content from DOCX bytes using python-docx"""
+    try:
+        from docx import Document
+        import io
+
+        docx_file = io.BytesIO(docx_bytes)
+        doc = Document(docx_file)
+
+        text_content = []
+        
+        # Extract text from paragraphs
+        for paragraph in doc.paragraphs:
+            if paragraph.text.strip():
+                text_content.append(paragraph.text)
+        
+        # Extract text from tables
+        for table in doc.tables:
+            for row in table.rows:
+                for cell in row.cells:
+                    if cell.text.strip():
+                        text_content.append(cell.text)
+
+        full_text = "\n".join(text_content).strip()
+        return full_text if full_text else "DOCX_NO_TEXT: No readable text content found in DOCX."
+
+    except ImportError:
+        return "DOCX_ERROR: python-docx library not available. Install with: pip install python-docx"
+    except Exception as e:
+        return f"DOCX_ERROR: Error reading DOCX content: {str(e)}"
 
 if len(sys.argv) < 4:
     print("Usage: python index_datasets.py <storage_account_name> <blob_container_name> <ai_search_endpoint> [<ai_search_index_name>]")
@@ -96,6 +129,8 @@ def extract_pdf_text(pdf_bytes):
     title = blob.name.replace(".csv", "")
     title = title.replace(".json", "")
     title = title.replace(".pdf", "")
+    title = title.replace(".docx", "")
+    title = title.replace(".pptx", "")
     data = container_client.download_blob(blob.name).readall()
 
     try:
@@ -104,6 +139,8 @@ def extract_pdf_text(pdf_bytes):
         # Check if this is a PDF file and process accordingly
         if blob.name.lower().endswith('.pdf'):
             text = extract_pdf_text(data)
+        elif blob.name.lower().endswith('.docx'):
+            text = extract_docx_text(data)
         else:
             # Original processing for non-PDF files
             text = data.decode('utf-8')