Skip to main content
Here you can find some exemples of analytical & agentic workflows that can be built & executed to analyse complex connected data with TuringDB
Unfortunately, the open source workflow LangGraph wrapper could not be added to the August2025 release in time, but it will be made available in the September2025 release. Right now the workflows can be used in our TuringDB cloud

​
Multi-Documents to Graph

​
See tutorial for hands on exemple: Clinical Notes - Multidocs tutorial

Here we take multiple clinical notes documents (for the exemple we are using synthetic data), to uncover relationships between treatments, markers, health events, symptoms, clinical outcomes of a patient across multiple encounters. 📔 Notebook exported as HTML
  • 👨🏻‍💻 Workflow code 
    %%time 

builder = FlowBuilder()

# S3ListFiles
files = builder.add_node(
    S3ListFiles.Node(
        params=S3ListFiles.Params(
            user_id=user_id,
            filenames=list_filenames,
            max_item_count=2
        )
    )
)

# InputText - Query input
input_query = builder.add_node(
    InputText.Node(
        params=InputText.Params(
            output_field="$query",
            content=read_query_str
        ),
    ),
)

# Merge
merge = builder.add_node(
    Merge.Node()
)

# ForEach
for_each = builder.add_node(
    ForEach.Node(
        params=ForEach.Params(
            collection_field="files",
            current_field="i"
        )
    )
)

# S3LoadFile
s3_loader = builder.add_node(
    S3LoadFile.Node(
        params=S3LoadFile.Params(
            output_field=DataField(field="$file_content", action="append"),
            
            user_id=user_id,
            file_key=DataField(field="$files[$i].key", action="set"),
            file_type="text",
        )
    )
)

# MistralEntityExtractor
extractor = builder.add_node(
    MistralEntityExtractor.Node(
        params=MistralEntityExtractor.Params(
            input_field="$file_content[$i].data",
            output_field=DataField(field="$list_entities", action="append"),
        )
    )
)

# MergeEntities
merge_entities = builder.add_node(
    MergeEntities.Node(
        params=MergeEntities.Params(
            input_field="$list_entities",
            output_field="merged_entities"
        )
    )
)

# Create graph
write_query = builder.add_node(
    TuringDBWrite.Node(
        params=TuringDBWrite.Params(
            input_field="merged_entities",
            instance_id=instance_id,
            auth_token=auth_token,
            graph_name=graph_name,
        )
    )
)

# Read graph
read_query = builder.add_node(
    TuringDBQuery.Node(
        params=TuringDBQuery.Params(
            input_field="query",
            output_field="query_result",
            instance_id=instance_id,
            auth_token=auth_token,
            graph_name=graph_name
        )
    )
)

# Generate GML
generator = builder.add_node(
    GmlGenerator.Node(
        params=GmlGenerator.Params(
            input_field="$merged_entities",
            output_field="gml"
        )
    )
)

# Graph summary
explainer = builder.add_node(
    MistralGraphExplainer.Node(
        params=MistralGraphExplainer.Params(
            input_field="gml",
            output_field="gml_summary"
        )
    )
)

# Text Output
out = builder.add_node(
    OutputText.Node(
        params=OutputText.Params(
            input_field="gml_summary",
            output_field="output"
        ),
    )
)

# Merge
merge2 = builder.add_node(
    Merge.Node()
)

# Connect nodes
files.connect_to(merge)
input_query.connect_to(merge)
merge.connect_to(for_each)

for_each.connect_to(s3_loader)
s3_loader.connect_to(extractor)
extractor.connect_to(for_each)
for_each.connect_to(merge_entities)
merge_entities.connect_to(write_query)
write_query.connect_to(read_query)
merge_entities.connect_to(generator)
generator.connect_to(explainer)
explainer.connect_to(out)

out.connect_to(merge2)
read_query.connect_to(merge2)

# Build flow
pipeline = builder.build()

# Show pipeline image
pipeline

​
Document to Graph

Here we study the supply chain of Apple trying to see the interaction and dependencies on different countries & regions for different parts (e.g. batteries, screens, etc) of Apple products. The reference document used is a report from the American Enterprise Institute 📔 Notebook exported as HTML
  • 👨🏻‍💻 Workflow code 
    %%time

builder = FlowBuilder()

# S3LoadFile node
s3_loader = builder.add_node(
    S3LoadFile.Node(
        params=S3LoadFile.Params(
            output_field="pdf_base64",
            
            user_id=user_id,
            file_key=file_key_pdf_input,
            file_type="pdf"
        )
    )
)

# ExtractTextPDF
pdf_text_extractor = builder.add_node(
    ExtractTextPDF.Node(
        name="ExtractTextPDF",
        params=ExtractTextPDF.Params(
            input_field="pdf_base64",
            output_field="pdf_text"
        )
    )
)

# Entity Extractor
extractor = builder.add_node(
    MistralEntityExtractor.Node(
        name="Entity Extractor",
        params=MistralEntityExtractor.Params(
            input_field="pdf_text.all_content",
            output_field="extracted_entities"
        )
    )
)

# GML Generator node
generator = builder.add_node(
    GmlGenerator.Node(
        name="GML Generator",
        params=GmlGenerator.Params(
            input_field="extracted_entities",
            output_field="gml_generated"
        )
    )
)

# Graph Explainer
explainer = builder.add_node(
    MistralGraphExplainer.Node(
        name="Graph Explainer",
        params=MistralGraphExplainer.Params(
            input_field="gml_generated",
            output_field="gml_summary"
        )
    )
)

# Text Output
out = builder.add_node(
    OutputText.Node(
        name="Text Output",
        params=OutputText.Params(
            input_field="gml_summary"
        ),
    )
)

# Connect nodes
s3_loader.connect_to(pdf_text_extractor)
pdf_text_extractor.connect_to(extractor)
extractor.connect_to(generator)
generator.connect_to(explainer)
explainer.connect_to(out)

# Build flow
pipeline = builder.build()

# Execute flow
results = pipeline.execute()

# Show pipeline image
pipeline

​
Database Integration

Integrate a simple relational database to be transformed in a graph database structure to visualise, analyse, and reason on the graphs. Example with our partner antibody database from CiteAb (sample database): From table structure to graph 📔 Notebook exported as HTML
  • 👨🏻‍💻 Workflow code 
    %%time

builder = FlowBuilder()

# Text Input - Gml 
input_gml = builder.add_node(
    InputText.Node(
        params=InputText.Params(
            output_field="gml_raw",
            content=gml_content
        ),
    ),
)

# Text Input - Query
input_query = builder.add_node(
    InputText.Node(
        params=InputText.Params(
            output_field="query_natural_language",
            content=query_natural_language
        ),
    ),
)

# Text Input - Graph entities
input_entities = builder.add_node(
    InputText.Node(
        params=InputText.Params(
            output_field="entities",
            content=entities_obj
        ),
    ),
)

# Merge
merge1 = builder.add_node(
    Merge.Node()
)

# Graph Explainer
explainer = builder.add_node(
    MistralGraphExplainer.Node(
        params=MistralGraphExplainer.Params(
            input_field="gml_raw",
            output_field="gml_summary"
        )
    )
)

# Text2Cypher
text2cypher = builder.add_node(
    Text2Cypher.Node(
        name="Text2Cypher",
        params=Text2Cypher.Params(
            input_field="query_natural_language",
            output_field="cypher_query",
            
            llm_provider="OpenAI",
            #api_key_openai=api_keys["OpenAI"],
            model="gpt-4.1",
            
            # Optional: Provide graph schema for better queries
            node_labels=node_labels,
            relationship_types=relationship_types, 
            property_keys=property_keys,

            # Optional: Provide graph raw gml content
            gml_content=gml_content,
            
            # Optional: Database description
            database_description="Graph containing interactions between antibodies and proteins (with parameters) from CiteAb life science data company",
        )
    )
)

# Create graph
write_query = builder.add_node(
    TuringDBWrite.Node(
        params=TuringDBWrite.Params(
            input_field="entities",
            instance_id=instance_id,
            auth_token=auth_token,
            graph_name=graph_name,
        )
    )
)

# Read graph
read_query = builder.add_node(
    TuringDBQuery.Node(
        params=TuringDBQuery.Params(
            input_field="cypher_query",
            output_field="result_query",
            instance_id=instance_id,
            auth_token=auth_token,
            graph_name=graph_name
        )
    )
)

# Merge
merge2 = builder.add_node(
    Merge.Node()
)

# Connect nodes
input_gml.connect_to(merge1)
input_query.connect_to(merge1)
input_entities.connect_to(merge1)

merge1.connect_to(explainer)
merge1.connect_to(text2cypher)

text2cypher.connect_to(write_query)
write_query.connect_to(read_query)

explainer.connect_to(merge2)
read_query.connect_to(merge2)

# Build flow
pipeline = builder.build()

# Show pipeline image
pipeline

​
Automatic Queries from APIs

Here we define a list of stock tickers from the ARKK ETF to automatically query them from a finance data API (i.e., Polygon or FinancialData) to then be analysed as a correlation graph of the stocks and as a dependency graph of the companies in the ARKK ETF portfolio. 📔 Notebook exported as HTML
  • 👨🏻‍💻 Workflow code 
    %%time

builder = FlowBuilder()

# S3LoadFile node
s3_loader = builder.add_node(
    S3LoadFile.Node(
        params=S3LoadFile.Params(
            output_field="pdf_base64",
            
            user_id=user_id,
            file_key=file_key_pdf_input,
            file_type="pdf"
        )
    )
)

# ExtractTextPDF
pdf_text_extractor = builder.add_node(
    ExtractTextPDF.Node(
        name="ExtractTextPDF",
        params=ExtractTextPDF.Params(
            input_field="pdf_base64",
            output_field="pdf_text"
        )
    )
)

# LLM
llm = builder.add_node(
    LLM.Node(
        name="LLM",
        params=LLM.Params(
            input_field="pdf_text.all_content",
            output_field="list_companies_tickers",
            api_key_anthropic="<your_api_key_anthropic>",
            llm_provider="Anthropic",
            system_prompt="""
            Your role :
            - Extract companies tickers from this input text
            - Return the extracted companies tickers as a list
            Very important :
            - Return only this list, no other explanations
            """,
            output_format="list"
        )
    )
)

# ForEach
for_each = builder.add_node(
    ForEach.Node(
        params=ForEach.Params(
            list_field="list_companies_tickers",
            offset_field="i"
        )
    )
)

# FinancialDataRestAPI : stock prices
financial_data = builder.add_node(
    FinancialDataRestAPI.Node(
        params=FinancialDataRestAPI.Params(
            output_field=DataField(field="$stock_prices", action="append"),
            
            api_key=financial_data_api_key,
            endpoint_type="stock_prices",
            identifier=DataField(field="$list_companies_tickers[$i]", action="set"),
            #offset=0,
            #format="json"
        )
    )
)

# OutputText - Get all data
out = builder.add_node(
    OutputText.Node(
        params=OutputText.Params(
            input_field="$stock_prices[$i]",
            output_field=DataField(field="$stock_prices_all[$i]", action="append")
        )
    )
)

# Connect nodes
s3_loader.connect_to(pdf_text_extractor)
pdf_text_extractor.connect_to(llm)
llm.connect_to(for_each)
for_each.connect_to(financial_data)
financial_data.connect_to(out)
out.connect_to(for_each)

# Build flow
pipeline = builder.build()

# Execute flow
results = pipeline.execute()

# Show pipeline image
pipeline
⌘I