STEP 1: Parallelized Data Collection

Supervisor Agent dispatches specialized agents to fetch relevant data in parallel.

GBQ Research Agent
- Fetches structured internal datasets from BigQuery.
DailyMed Agent
- Retrieves verified drug information and monographs.
Web Research Agent
- Crawls and scrapes open medical research and news.

STEP 2: Data Synthesis & Gap Analysis

The Synthesizer Agent merges the raw findings into a preliminary brief, resolving conflicts and identifying gaps.

Input
- Data from all Step 1 Agents.
⟶
Synthesizer Agent
- Creates: Preliminary Research Brief
Highlights inconsistencies and missing data points.

STEP 3: Collaborative Summarization & Voting

Multiple LLM agents draft independent summaries, then vote to reach a unified consensus.

Summarizer Agent A
- Powered by Claude
- Drafts version 1
Summarizer Agent B
- Powered by Gemini
- Drafts version 2
Summarizer Agent C
- Powered by GPT-4o
- Drafts version 3
⟶
Critique & Voting Agent
- Compares drafts against a rubric
- Creates: Consensus Summary
Integrates strengths from all drafts.

STEP 4: Human-in-the-Loop Review

Medical researchers review AI-generated content, provide feedback, and request clarification.

Interface Viewer
- Displays Consensus Summary and source info
- Enables targeted review by human experts
⟶
Medical Researcher (MR)
- Reviews key chunks
- Provides corrections, queries, and follow-up requests

STEP 5: Final Report Generation

The final LLM synthesizes validated data into a formal, structured report.

Report Generation Agent
- Powered by GPT-4o / Gemini 1.5 Pro
- Inputs: raw data, consensus summary, MR feedback
⟶
Final Research Document
- Fully structured, validated, and ready for submission or publication