Parallel Sub-Agents Research Pipeline

A research assistant built with LangGraph that fans out to three parallel sub-agents (Knowledge Base, Web Search, Policy), aggregates results using state reducers, and synthesizes a single customer-facing response. Includes a conditional routing variant and LangSmith evaluations.

Prerequisites

• Python 3.11+
• Anthropic API key
• LangSmith API key (for evals)

Setup

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Environment Variables

Variable	Required	Description
ANTHROPIC_API_KEY	Yes	Anthropic API key for Claude
LANGSMITH_API_KEY	For evals	LangSmith API key for tracing and evals
LANGSMITH_TRACING	No	Set to true to enable tracing

Running

Run the basic parallel agent (static fan-out to all three agents):

python parallel_agents.py

Run the conditional routing variant (classifies query, dispatches only relevant agents):

python conditional_routing.py

Run evaluations:

python evals.py

Files

File	Description
parallel_agents.py	Static fan-out to KB, Web, and Policy agents with operator.add reducer and synthesis
conditional_routing.py	Structured-output classifier that selectively dispatches agents based on query type
evals.py	LangSmith evaluation suite: coverage, source diversity, and LLM-as-judge quality

Article

Your Customer Service Bot Is Slow Because It's Single-Threaded