In this example, we’ll build a blog content generation Workflow that creates engaging, targeted content through a series of prompts. The Workflow takes an industry and target audience as inputs and produces a complete blog post through the following steps:

TopicResearch: Generates and rates potential blog topics based on audience relevance and SEO potential
OutlineCreation: Creates a detailed outline for the highest-rated topic
ContentGeneration: Writes a complete blog post following the outline
FinalOutput: Returns the final blog content

workflow.py

1 ## Graph Definition
2 class Workflow(BaseWorkflow[Inputs, BaseState]):
3     graph = TopicResearch >> OutlineCreation >> ContentGeneration >> FinalOutput
4 
5     class Outputs(BaseWorkflow.Outputs):
6         final_output = FinalOutput.Outputs.value
7 
8 ## Running it
9 workflow = Workflow()
10 terminal_event = workflow.run(
11     inputs=Inputs(
12         audience="Working professionals targeting midsize businesses",
13         industry="Marketing Technology"
14     )
15 )
16 
17 ## Output:
18 print(terminal_event.outputs.final_output)
19 
20 """
21 [Output will be a complete blog post tailored to the specified industry and audience]
22 """

Which corresponds to a Workflow graph like this:

Setup

Install Vellum

$ pip install vellum-ai

Create your Project

Structure your project like this:

1 basic_prompt_chain/
2 ├── workflow.py
3 ├── inputs.py
4 ├── sandbox.py
5 ├── __init__.py
6 └── nodes/
7     ├── __init__.py
8     ├── topic_research.py
9     ├── outline_creation.py
10     ├── content_generation.py
11     └── final_output.py

Define Workflow Inputs

inputs.py

1 from vellum.workflows.inputs import BaseInputs
2 
3 class Inputs(BaseInputs):
4     audience: str
5     industry: str

The Workflow takes two inputs: the target audience and industry. These will be used to generate relevant, targeted content throughout the chain.

Build the Nodes

Topic Research

This node generates and rates potential blog topics based on the specified industry and audience.

nodes/topic_research.py

1 class TopicResearch(InlinePromptNode):
2     ml_model = "gpt-4o-mini"
3     blocks = [
4         ChatMessagePromptBlock(
5             chat_role="SYSTEM",
6             blocks=[
7                 JinjaPromptBlock(
8                     template="""\
9 You are a content strategist who identifies engaging blog topics for businesses to help boost their online presence and SEO performance. 
10 
11 You will suggest 3 blog topics for:
12 Industry: {{ industry }}
13 Target Audience: {{ audience }}
14 
15 For each topic, provide:
16 1. A compelling title
17 2. A brief description of the angle
18 3. A rating from 1-10 with explanation
19 
20 No preamble/postamble\
21 """
22                 )
23             ],
24         ),
25     ]
26     prompt_inputs = {
27         "audience": Inputs.audience,
28         "industry": Inputs.industry,
29     }

Outline Creation

Takes the rated topics and creates a detailed outline for the highest-rated one.

nodes/outline_creation.py

1 class OutlineCreation(InlinePromptNode):
2     ml_model = "gpt-4o-mini"
3     blocks = [
4         ChatMessagePromptBlock(
5             chat_role="SYSTEM",
6             blocks=[
7                 JinjaPromptBlock(
8                     template="""\
9 You are a content outline specialist. 
10 You will receive a list of <topics> with ratings to create a detailed outline targeted at {{ audience }}.
11 
12 You will:
13 1. Select the highest-rated topic(s)
14 2. Create a detailed outline for a 1000-word blog post
15 3. Include sections for examples and actionable tips
16 4. Add placeholders for statistics or case studies
17 
18 No preamble/postamble\
19 """
20                 )
21             ],
22         ),
23     ]
24     prompt_inputs = {
25         "audience": Inputs.audience,
26         "topics": TopicResearch.Outputs.text,
27     }

Content Generation

Transforms the outline into a complete blog post.

nodes/content_generation.py

1 class ContentGeneration(InlinePromptNode):
2     ml_model = "gpt-4o-mini"
3     blocks = [
4         ChatMessagePromptBlock(
5             chat_role="SYSTEM",
6             blocks=[
7                 JinjaPromptBlock(
8                     template="""\
9 You are an expert content writer specializing in {{ industry }} for {{ audience }}.
10 
11 Guidelines:
12 1. Select the highest-rated topic
13 2. Write in a conversational tone
14 3. Follow the outline exactly
15 4. Include examples and statistics
16 5. Break up text with subheadings
17 6. Target length: 600 words
18 7. Include introduction and call-to-action
19 
20 No preamble/postamble\
21 """
22                 )
23             ],
24         ),
25     ]
26     prompt_inputs = {
27         "outline": OutlineCreation.Outputs.text,
28         "industry": Inputs.industry,
29         "audience": Inputs.audience,
30     }

Final Output

Returns the generated content as the Workflow output.

nodes/final_output.py

1 class FinalOutput(FinalOutputNode[BaseState, str]):
2     class Outputs(FinalOutputNode.Outputs):
3         value = ContentGeneration.Outputs.text

Instantiate the Graph and Invoke it

Define the Graph and its Outputs

workflow.py

1 from vellum.workflows import BaseWorkflow
2 from vellum.workflows.state import BaseState
3 
4 from .inputs import Inputs
5 from .nodes.topic_research import TopicResearch
6 from .nodes.outline_creation import OutlineCreation
7 from .nodes.content_generation import ContentGeneration
8 from .nodes.final_output import FinalOutput
9 
10 class Workflow(BaseWorkflow[Inputs, BaseState]):
11     graph = TopicResearch >> OutlineCreation >> ContentGeneration >> FinalOutput
12 
13     class Outputs(BaseWorkflow.Outputs):
14         final_output = FinalOutput.Outputs.value

You can output directly from the ContentGeneration node. Here we’ve used the FinalOutput node for clarity when pushed to the UI.

Instantiate the Workflow

1 ## From any file / function from which you want to reference the Workflow
2 
3 # Required import (the file imported from depends on your folder structure)
4 # from .workflow import Workflow
5 
6 workflow = Workflow()

Invoke the Workflow

1 ## From any file / function from which you want to run the Workflow
2 
3 # Required imports (the file imported from depends on your folder structure)
4 # from .inputs import Inputs
5 
6 terminal_event = workflow.run(
7     inputs=Inputs(
8         audience="Working professionals targeting midsize businesses",
9         industry="Marketing Technology"
10     )
11 )
12 
13 ## Get the generated content:
14 print(terminal_event.outputs.final_output)
15 
16 """
17 Example Output:
18 5 Essential MarTech Stack Optimization Strategies for 2024
19 
20 In today's rapidly evolving business landscape, having an efficient marketing technology stack is no longer optional—it's crucial for success. As marketing professionals targeting midsize businesses, you need to ensure your MarTech investments deliver maximum ROI while staying within budget constraints...
21 
22 [Rest of generated blog post content...]
23 """

Conclusion

We’ve built a prompt chain Workflow that generates blog content through a series of specialized prompts. Each node in the chain builds upon the previous one’s output, resulting in high-quality, targeted content. From here, you can:

Version control the Workflow with your codebase
Continue development in the Vellum UI
Deploy to Vellum
Add evaluation metrics to assess content quality
Expand the chain with additional steps like SEO optimization or image generation

1	## Graph Definition
2	class Workflow(BaseWorkflow[Inputs, BaseState]):
3	graph = TopicResearch >> OutlineCreation >> ContentGeneration >> FinalOutput
4
5	class Outputs(BaseWorkflow.Outputs):
6	final_output = FinalOutput.Outputs.value
7
8	## Running it
9	workflow = Workflow()
10	terminal_event = workflow.run(
11	inputs=Inputs(
12	audience="Working professionals targeting midsize businesses",
13	industry="Marketing Technology"
14	)
15	)
16
17	## Output:
18	print(terminal_event.outputs.final_output)
19
20	"""
21	[Output will be a complete blog post tailored to the specified industry and audience]
22	"""

1	basic_prompt_chain/
2	├── workflow.py
3	├── inputs.py
4	├── sandbox.py
5	├── __init__.py
6	└── nodes/
7	├── __init__.py
8	├── topic_research.py
9	├── outline_creation.py
10	├── content_generation.py
11	└── final_output.py

1	from vellum.workflows.inputs import BaseInputs
2
3	class Inputs(BaseInputs):
4	audience: str
5	industry: str

1	class TopicResearch(InlinePromptNode):
2	ml_model = "gpt-4o-mini"
3	blocks = [
4	ChatMessagePromptBlock(
5	chat_role="SYSTEM",
6	blocks=[
7	JinjaPromptBlock(
8	template="""\
9	You are a content strategist who identifies engaging blog topics for businesses to help boost their online presence and SEO performance.
10
11	You will suggest 3 blog topics for:
12	Industry: {{ industry }}
13	Target Audience: {{ audience }}
14
15	For each topic, provide:
16	1. A compelling title
17	2. A brief description of the angle
18	3. A rating from 1-10 with explanation
19
20	No preamble/postamble\
21	"""
22	)
23	],
24	),
25	]
26	prompt_inputs = {
27	"audience": Inputs.audience,
28	"industry": Inputs.industry,
29	}

1	class OutlineCreation(InlinePromptNode):
2	ml_model = "gpt-4o-mini"
3	blocks = [
4	ChatMessagePromptBlock(
5	chat_role="SYSTEM",
6	blocks=[
7	JinjaPromptBlock(
8	template="""\
9	You are a content outline specialist.
10	You will receive a list of <topics> with ratings to create a detailed outline targeted at {{ audience }}.
11
12	You will:
13	1. Select the highest-rated topic(s)
14	2. Create a detailed outline for a 1000-word blog post
15	3. Include sections for examples and actionable tips
16	4. Add placeholders for statistics or case studies
17
18	No preamble/postamble\
19	"""
20	)
21	],
22	),
23	]
24	prompt_inputs = {
25	"audience": Inputs.audience,
26	"topics": TopicResearch.Outputs.text,
27	}

1	class ContentGeneration(InlinePromptNode):
2	ml_model = "gpt-4o-mini"
3	blocks = [
4	ChatMessagePromptBlock(
5	chat_role="SYSTEM",
6	blocks=[
7	JinjaPromptBlock(
8	template="""\
9	You are an expert content writer specializing in {{ industry }} for {{ audience }}.
10
11	Guidelines:
12	1. Select the highest-rated topic
13	2. Write in a conversational tone
14	3. Follow the outline exactly
15	4. Include examples and statistics
16	5. Break up text with subheadings
17	6. Target length: 600 words
18	7. Include introduction and call-to-action
19
20	No preamble/postamble\
21	"""
22	)
23	],
24	),
25	]
26	prompt_inputs = {
27	"outline": OutlineCreation.Outputs.text,
28	"industry": Inputs.industry,
29	"audience": Inputs.audience,
30	}

1	class FinalOutput(FinalOutputNode[BaseState, str]):
2	class Outputs(FinalOutputNode.Outputs):
3	value = ContentGeneration.Outputs.text

1	from vellum.workflows import BaseWorkflow
2	from vellum.workflows.state import BaseState
3
4	from .inputs import Inputs
5	from .nodes.topic_research import TopicResearch
6	from .nodes.outline_creation import OutlineCreation
7	from .nodes.content_generation import ContentGeneration
8	from .nodes.final_output import FinalOutput
9
10	class Workflow(BaseWorkflow[Inputs, BaseState]):
11	graph = TopicResearch >> OutlineCreation >> ContentGeneration >> FinalOutput
12
13	class Outputs(BaseWorkflow.Outputs):
14	final_output = FinalOutput.Outputs.value

1	## From any file / function from which you want to reference the Workflow
2
3	# Required import (the file imported from depends on your folder structure)
4	# from .workflow import Workflow
5
6	workflow = Workflow()

1	## From any file / function from which you want to run the Workflow
2
3	# Required imports (the file imported from depends on your folder structure)
4	# from .inputs import Inputs
5
6	terminal_event = workflow.run(
7	inputs=Inputs(
8	audience="Working professionals targeting midsize businesses",
9	industry="Marketing Technology"
10	)
11	)
12
13	## Get the generated content:
14	print(terminal_event.outputs.final_output)
15
16	"""
17	Example Output:
18	5 Essential MarTech Stack Optimization Strategies for 2024
19
20	In today's rapidly evolving business landscape, having an efficient marketing technology stack is no longer optional—it's crucial for success. As marketing professionals targeting midsize businesses, you need to ensure your MarTech investments deliver maximum ROI while staying within budget constraints...
21
22	[Rest of generated blog post content...]
23	"""