CAG Deep Research

Multi-agent research automation with LangGraph orchestration.

LangGraphMulti-AgentRAGOllama

Code

Agent Pipeline

Query

Analyze

Verify

Synthesize

Report

Query → 5+ specialized agents → Comprehensive research report

Specialized Agents

Orchestrator

Decomposes queries

Web Searcher

Finds sources

Analyzer

Extracts info

Fact Checker

Validates claims

Synthesizer

Combines findings

Editor

Refines output

Hexagonal Architecture

Domain (Core)

Research WorkflowAgent CoordinationState Machine

Ports (Interfaces)

LLMPortSearchPortStoragePort

Adapters (Impl)

OllamaOpenAITavilyPostgreSQL

Hybrid LLM Strategy

Local (Ollama)

Llama 3, Mistral

Content analysis, summarization, extraction — commodity tasks at 1/10th cost

Cloud (Fallback)

GPT-4, Claude

Fact-checking, synthesis, complex reasoning — premium tasks requiring quality

Specialized Agents

50+

Sources per Query

~70%

Cost Reduction

<15m

Full Research Cycle

"Multi-agent systems excel when tasks have natural decomposition boundaries. Research has clear phases—search, analyze, verify, synthesize—that map perfectly to specialized agents."