Mercury Intelligence / Architecture — Web-Augmented
Yeo Zhi Yong
Web-Augmented

Web-Augmented · Live Search Before Reasoning

Two parallel web searches inject live context before the model reasons. Same output schema — completely different quality of intelligence.

⬆ Adds over Single Prompt: Live web retrieval — real-time search results are structured into the prompt before the model reasons, grounding output in current events rather than training knowledge alone.
👤
User
Mercury Intelligence
Input
company name
⚙️
Flask API
POST /api/brief/v2
Orchestration
2 search queries
New
🔍
Web Search
Live results
Retrieval
top 6 results + URLs
🧠
Claude Sonnet
Training + live context
Reasoning
JSON + sources
📋
Account Brief
Brief + _sources array
Output
1
User submits a company name via UI or API
2
Two web searches run in parallel — latest news + technology strategy
3
Top 6 results are injected into the model prompt as structured context
4
Model reasons with live data; brief returns with source URLs for transparency
Mercury Intelligence · Yeo Zhi Yong
⚡ Gap: fixed search queries — autonomous strategy fixed by Agentic Loop