Sub-Question Decomposition Strategy
For complex multi-hop questions, split into sub-questions, answer each, then synthesize a final answer.
Sub-Question Decomposition Strategy is a free AI Agents lesson on CoddyKit — lesson 4 of 4. You can read the complete lesson below for free — then practise it hands-on in the browser with a built-in code editor and a 24/7 AI tutor. It is part of the AI Agents learning path, one of 4 lessons in the course, and your progress syncs across the web and the CoddyKit app.
When Single-Hop Fails
Some questions need multiple lookups:
"Compare the revenue of Apple in 2022 and Microsoft in 2023."
One similarity search will not retrieve both facts; you need TWO sub-queries.
Sub-Question Decomposition
LlamaIndex SubQuestionQueryEngine splits a question into sub-questions, answers each, then synthesises:
from llama_index.core.query_engine import SubQuestionQueryEngine
from llama_index.core.tools import QueryEngineTool, ToolMetadata
query_engine_tools = [
QueryEngineTool(
query_engine=apple_index.as_query_engine(),
metadata=ToolMetadata(name='apple', description='Financial data for Apple Inc.')
),
QueryEngineTool(
query_engine=msft_index.as_query_engine(),
metadata=ToolMetadata(name='microsoft', description='Financial data for Microsoft Corp.')
)
]
engine = SubQuestionQueryEngine.from_defaults(query_engine_tools=query_engine_tools)
response = engine.query('Compare 2022 revenue of Apple vs 2023 revenue of Microsoft')How It Works
- LLM analyses the question
- LLM generates sub-questions and assigns each to a tool
- Each sub-question runs in parallel
- LLM synthesises a final answer from the sub-answers
Sub-Question Visibility
You can see the sub-questions the engine generated:
response = engine.query('Compare X and Y')
for sq in response.metadata.get('sub_qa', []):
print(sq.sub_q.sub_question, '->', sq.answer)When to Use Sub-Question
- Comparison queries (X vs Y)
- Multi-source aggregation
- Questions with multiple distinct facts
When NOT to Use
- Simple Q&A — adds latency and cost
- Single-document context
- Tasks with no decomposition value
Routing With LLM
Sub-Question Engine pairs with RouterQueryEngine to first decide which collection of indexes to even hit:
from llama_index.core.query_engine import RouterQueryEngine
router = RouterQueryEngine.from_defaults(query_engine_tools=query_engine_tools)
# Routes 'apple revenue' to apple index, 'msft margins' to msft, etc.Sub-Questions for Coding Agents
For agents that write code, sub-question is a way to plan steps:
"Build a REST API" -> sub: design models, sub: write CRUD handlers, sub: add auth.
Cost Implications
Sub-question makes N sub-queries instead of 1. Each is an LLM call + retrieval. Budget accordingly — 3-4x more expensive than single-hop.
Compared to ReAct
Sub-question is upfront planning. ReAct (next course) is step-by-step. For known-decomposable queries, sub-question is faster; for exploratory tasks, ReAct adapts better.
Combined: ReAct + Sub-Question Tools
The most flexible: a ReAct agent where sub-question engines are tools — each "tool call" can itself decompose into many sub-queries.
Eval Sub-Question Quality
Track whether the LLM generates sensible sub-questions. Bad decomposition is hard to debug without inspecting traces.
Decomposition Use Case
For which kind of question is sub-question decomposition most useful?
Recap
Sub-question decomposition turns one complex query into several simple ones — at higher cost but much higher accuracy on comparison and multi-source questions.
Frequently asked questions
Is the “Sub-Question Decomposition Strategy” lesson free?
Yes — the full text of “Sub-Question Decomposition Strategy” is free to read here on the web, and the AI Agents course includes 4 lessons in total. To practise it interactively (a built-in code editor and a 24/7 AI tutor) and unlock the rest of the AI Agents course, upgrade to CoddyKit PRO.
What will I learn in “Sub-Question Decomposition Strategy”?
For complex multi-hop questions, split into sub-questions, answer each, then synthesize a final answer. You practise AI Agents with hands-on code you run directly in the browser, and a 24/7 AI tutor answers your questions as you work through the lesson.
Do I need any experience to start AI Agents?
No prior experience is required. AI Agents on CoddyKit is structured for beginners through advanced learners; this is — lesson 4 of 4, so you can start here or from the beginning and move at your own pace.
How long does the “Sub-Question Decomposition Strategy” lesson take?
Most CoddyKit lessons take about 5–10 minutes. Each one is bite-sized and interactive, so you make steady progress and pick up exactly where you left off across the web and the app.
Can I write and run code in this AI Agents lesson?
Yes. Every AI Agents lesson includes a built-in code editor, so you write and run real code right in your browser and get instant AI feedback — no local setup required.
All lessons in this course
- Document Loaders and Parsers
- The Index Hierarchy: Vector, Tree, Keyword
- Query Engines and Response Synthesis
- Sub-Question Decomposition Strategy