How Iris Works¶

# Import necessary modules
from tulip.core.collection import TulipCollection
from tulip.data.haver import HaverClient as hc

# Display all outputs in cells
from IPython.core.interactiveshell import InteractiveShell
from tulip.data.bloomberg.collections import US_WEEKLY_ACTIVITY
from tulip.data.bloomberg import BloombergClient as bb
from tulip.data.fred import get_series_from_fred
from IPython.display import display, Markdown, HTML

InteractiveShell.ast_node_interactivity = "all"

from tulip.genai import iris

We first create a collection of time-series we want to analyze:

wei_collection = bb.create_collection(US_WEEKLY_ACTIVITY)
wei_collection["INJCJC   Index"].good_is = -1
wei_collection["INJCUIFE Index"].good_is = -1
wei_collection.dashboard.table()
wei_collection.via = "bloomberg"

wei_collection.dashboard.plots(years_limit=3, show_0=True)

We then pose the question to Iris:

question = iris.summarize(
    wei_collection,
    bullets=3,
    custom_prompt="This is Weekly Economic Indicator data. What trends can you see? Not all these time series are equally important. For example, the WEI is an aggregation of many time-series that are underneath"
    "so you should not overindex to it. Sales and Employment data are more important than noisy measures like US electricity output",
)

And print it outputmm

question.html()

data_input = question.data_input
data_input = data_input.replace("\\n", "\n")
display(Markdown(f"```text\n{data_input}\n```"))

display(Markdown(question.prompt))

Using an alternative model:

question = iris.summarize(
    wei_collection,
    bullets=5,
    model="claude-sonnet-4-20250514",
    custom_prompt="This is Weekly Economic Indicator data. What trends can you see? Not all these time series are equally important. For example, the WEI is an aggregation of many time-series that are underneath"
    "so you should not overindex to it. Sales and Employment data are more important than noisy measures like US electricity output",
)

question.html()