rag-search

name	rag-search
description	Search the internal knowledge base for policies, regulations, product documentation, and CIO investment views — account opening, KYC, mortgages, investment funds, ETFs, FINMA guidelines, and house investment recommendations
enabled	true
index_name	wm-knowledge-base

Instructions

When the user asks about internal policies, regulatory requirements, product documentation, CIO investment views, or compliance topics, use the rag_search tool to retrieve relevant content from the Azure AI Search knowledge base.

IMPORTANT: Always pass index_name: "wm-knowledge-base" when calling rag_search. This is the dedicated search index for the wealth-management knowledge base.

1. Determine When to Use

Use this skill when the user's question relates to any of the following domains:

Domain	Example queries
Account opening	"What documents are required to open a new account?", "What's the onboarding process for a corporate client?"
KYC / AML review	"What are the KYC requirements for high-risk clients?", "When is enhanced due diligence required?", "PEP screening procedures"
Mortgage policies	"What are the current mortgage lending criteria?", "LTV limits for residential mortgages", "Mortgage approval workflow"
Investment funds & ETFs	"What ETFs are available in the balanced strategy?", "Fund factsheet for emerging market equity", "Minimum investment thresholds"
FINMA regulations	"FINMA guidelines on client classification", "Cross-border advisory rules", "FINMA circular on outsourcing", "Regulatory capital requirements"
CIO recommendations & house views	"What is the bank's view on equities?", "CIO outlook on AI investments", "House recommendation on currencies and commodities", "Where does the CIO see growth opportunities?", "Investment themes for this quarter"
Investment research	"Opportunities in emerging markets", "Thematic investment views on technology", "Fixed income outlook", "Asset allocation recommendations"
Internal procedures	"Escalation process for suspicious transactions", "Client complaint handling", "Data retention policy"

Do NOT use this skill for:

• Real-time market data or current prices → use web_search
• Client-specific data (portfolio, profile) → use crm functions
• Calculations or analysis → use code_interpreter

2. Craft the Search Query

For best results with semantic search:

• Be specific — include the key regulatory or policy topic (e.g. "KYC enhanced due diligence requirements" rather than just "KYC")
• Include domain terms — use domain-specific language the knowledge base documents are likely to contain (e.g. "FINMA circular", "account opening checklist", "mortgage LTV ratio")
• For CIO / investment views — use terms like "opportunities", "outlook", "growth", "recommendation", "asset allocation", "thematic", combined with the asset class or sector (e.g. "opportunities in AI investments", "currency and commodity outlook", "seeking growth in equities")
• Rephrase if needed — if the first query returns low-relevance results, try an alternative phrasing or narrower scope

Quick-reference query examples:

Topic	Example queries
CIO views	"CIO outlook technology sector 2026", "house view emerging markets", "recommended allocation defensive portfolio"
KYC / AML	"KYC documentation requirements PEP", "AML enhanced due diligence thresholds"
Mortgages	"mortgage LTV limits lending criteria", "residential mortgage affordability calculation"
Funds / ETFs	"ETF selection criteria equity exposure", "fund of funds fee structure"
FINMA	"FINMA circular suitability requirements", "cross-border advisory restrictions"

3. Interpret Search Results

The rag_search tool returns a list of results, each containing:

• title — source document name
• content — relevant text excerpt (up to 500 characters)
• source — PDF filename
• page — page number in the original document
• score — relevance score from semantic ranking

When presenting results:

• Synthesize — combine information from multiple results into a coherent answer rather than listing raw excerpts
• Cite sources — reference the document title and page number so the user can verify (e.g. "According to Account Opening Policy (p. 12)...")
• Flag gaps — if the search returns no relevant results or low-confidence matches, tell the user and suggest they check with the compliance team or provide a more specific query
• Quote key passages — for regulatory or policy content, quote exact wording where precision matters (e.g. specific thresholds, deadlines, or prohibitions)

4. Multi-Step Workflows

Combine RAG search with other skills for richer answers:

• Policy + client context: Use rag_search to find the KYC policy → use crm to check the client's PEP status and documents → assess compliance
• Policy + summary: Use rag_search to retrieve a lengthy policy document → use document_summary to condense it
• Regulation + analysis: Use rag_search for FINMA guidelines on portfolio limits → use code_interpreter to check a client's portfolio against those limits
• Policy + email: Use rag_search to find the relevant procedure → use email_draft to draft a compliance notification
• CIO views + client portfolio: Use rag_search to retrieve the latest CIO recommendations or investment themes → use crm to get the client's portfolio → use code_interpreter to assess alignment between the house view and the client's current holdings → suggest rebalancing actions
• Investment research + report: Use rag_search to pull CIO outlook and thematic views → use pdf-wealth-report to generate a client-facing market update or investment proposal

5. Response Guidelines

• Present policy information clearly and accurately — do not paraphrase in ways that change the meaning
• Use bullet points or tables for checklists, requirements lists, and thresholds
• Always include the source document and page reference
• For regulatory content, note the date or version if visible in the document
• If multiple policies apply (e.g. both internal policy and FINMA regulation), present both and note any differences
• Remind the user that retrieved information should be verified with the compliance team for critical decisions