Mercury Intelligence / Architecture — Domain-Augmented Agent
Yeo Zhi Yong
Domain-Augmented Agent

Domain-Augmented Agent · Knowledge Base

The agent loop gains a domain knowledge base — MEDDPICC qualification criteria, Cisco APJC portfolio, and regional market intelligence. The model searches the KB alongside the web, grounding every brief in Cisco's own sales methodology.

⬆ Adds over Agentic Loop: Domain knowledge base — the agent now retrieves internal Cisco context (MEDDPICC, product portfolio, APJC market intel) via BM25 retrieval, not just public web data. Briefs become Cisco-specific and methodology-grounded.
👤
User
Mercury Intelligence
Input
company name
⚙️
Flask API
POST /api/brief/v4
Orchestration
tools + prompt
Domain-Augmented Agent Loop
KB
🧠
Claude Sonnet
Autonomous reasoning
+ 5 tools
call
result
External · Web Search
🔍
web_search
General search
📰
news_search
Recent news
💰
financials
Revenue · headcount
call
result
Internal · BM25
📚
search_knowledge_base
MEDDPICC
Cisco Portfolio
APJC Market Intel
BM25 · local
↺ loops until finish() called · max 8 rounds · KB via BM25
finish(brief) + sources + search log
👤
User
Mercury Intelligence
Receives brief
brief + MEDDPICC + cisco fit
📋
Account Brief v4
+ meddpicc_notes
+ cisco_fit
+ sources + log
Output
1
User submits a company name — model receives task + 5 tool schemas
2
Model runs web, news, and financial searches in parallel for live data
3
Model queries the KB — retrieves MEDDPICC criteria, Cisco products, APJC market context
4
Model calls finish() with a Cisco-grounded brief including MEDDPICC notes and product fit
Mercury Intelligence · Yeo Zhi Yong
✦ Next: AI governance layer