, the usual “textual content in, textual content out” paradigm will solely take you up to now.
Actual purposes that ship precise worth ought to have the ability to look at visuals, motive via complicated issues, and produce outcomes that programs can really use.
On this submit, we’ll design this stack by bringing collectively three highly effective capabilities: multimodal enter, reasoning, and structured output.
As an example this, we’ll stroll via a hands-on instance: constructing a time-series anomaly detection system for e-commerce order information utilizing OpenAI’s o3 mannequin. Particularly, we’ll present the way to pair o3’s reasoning functionality with picture enter and emit validated JSON, in order that the downstream system can simply eat it.
By the tip, our app will:
- See: analyze charts of e-commerce order quantity time collection
- Suppose: determine uncommon patterns
- Combine: output a structured anomaly report
You’ll depart with useful code you may reuse for numerous use instances that transcend simply anomaly detection.
Let’s dive in.
Focused on studying the broader panorama of how LLMs are being utilized for anomaly detection? Try my earlier submit: Boosting Your Anomaly Detection With LLMs, the place I summarized 7 rising utility patterns that you simply shouldn’t miss.
1. Case Research
On this submit, we intention to construct an anomaly detection resolution for figuring out irregular patterns in e-commerce order time collection information.
For this case examine, we generated three units of artificial every day order information. The datasets signify three totally different profiles of the every day order over roughly one month of time. To make seasonality apparent, we now have shaded the weekends. The x-axis exhibits the day of the week.
Every determine comprises one particular sort of anomaly (can you discover them?). We’ll later use these figures to check our anomaly detection resolution and see if it may well precisely get better these anomalies.
2. Our Answer
2.1 Overview
In contrast to the normal machine studying approaches that require tedious characteristic engineering and mannequin coaching, our present method is way easier. It really works with the next steps:
- We put together the determine for visualizing the e-commerce order time collection information.
- We immediate the reasoning mannequin o3, ask it to take a better have a look at the time collection picture we fed to it, and decide if an uncommon sample exists.
- The o3 mannequin will then output its findings in a pre-defined JSON format.
And that’s it. Easy.
In fact, to ship this resolution, we have to allow o3 mannequin to take picture enter and emit structured output. We are going to see how to try this shortly.
2.2 Organising the reasoning mannequin
As talked about earlier than, we’ll use o3 mannequin, which is the flagship reasoning mannequin from OpenAI that may deal with complicated multi-step issues with state-of-the-art efficiency. Particularly, we’ll use the Azure OpenAI endpoint to name the mannequin.
Ensure you have put the endpoint, API key, and deployment title in an .env file, we will then proceed to organising the LLM shopper:
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import matplotlib.dates as mdates
from openai import AzureOpenAI
from dotenv import load_dotenv
import os
load_dotenv()
# Setup LLM shopper
endpoint = os.getenv("api_base")
api_key = os.getenv("o3_API_KEY")
api_version = "2025-04-01-preview"
model_name = "o3"
deployment = os.getenv("deployment_name")
LLM_client = AzureOpenAI(
api_key=api_key,
api_version=api_version,
azure_endpoint=endpoint
)We use the next instruction because the system message for the o3 mannequin (tuned by GPT-5):
instruction = f"""
[Role]
You're a meticulous information analyst.
[Task]
You can be given a line chart picture associated to every day e-commerce orders.
Your activity is to determine outstanding anomalies within the information.
[Rules]
The anomaly varieties will be spike, drop, level_shift, or seasonal_outlier.
A level_shift is a sustained baseline change (≥ 5 consecutive days), not a single level.
A seasonal_outlier occurs if a weekend/weekday behaves not like friends in its class.
For instance, weekend orders are often decrease than the weekdays'.
Learn dates/values from axes; when you can’t learn precisely, snap to the closest tick and observe uncertainty in clarification.
The weekends are shaded within the determine.
"""Within the above instruction, we clearly outlined the function of the LLM, the duty that the LLM ought to full, and the foundations the LLM ought to comply with.
To restrict the complexity of our case examine, we deliberately specified solely 4 anomaly sorts that LLM must determine. We additionally offered clear definitions of these anomaly sorts to take away ambiguity.
Lastly, we injected a little bit of area data about e-commerce patterns, i.e., decrease weekend orders are anticipated in comparison with weekdays. Incorporating area know-how is usually thought-about good observe for guiding the mannequin’s analytical course of.
Now that we now have our mannequin arrange, let’s talk about the way to put together the picture for o3 mannequin to eat.
2.3 Picture preparation
To allow o3’s multimodal capabilities, we have to present figures in a particular format, i.e., both publicly accessible internet URLs or as base64-encoded information URLs. Since our figures are generated domestically, we’ll use the second method.
What’s Base64 Encoding anyway? Base64 is a option to signify binary information (like our picture information) utilizing solely textual content characters which might be protected to transmit over the web. It converts binary picture information right into a string of letters, numbers, and some symbols.
And what about information URL? A knowledge URL is a sort of URL that embeds the file content material instantly within the URL string, quite than pointing to a file location.
We will use the next perform to deal with this conversion robotically:
import io
import base64
def fig_to_data_url(fig, fmt="png"):
"""
Converts a Matplotlib determine to a base64 information URL with out saving to disk.
Args:
-----
fig (matplotlib.determine.Determine): The determine to transform.
fmt (str): The format of the picture ("png", "jpeg", and so forth.)
Returns:
--------
str: The information URL representing the determine.
"""
buf = io.BytesIO()
fig.savefig(buf, format=fmt, bbox_inches="tight")
buf.search(0)
base64_encoded_data = base64.b64encode(buf.learn()).decode("utf-8")
mime_type = f"picture/{fmt.decrease()}"
return f"information:{mime_type};base64,{base64_encoded_data}"Primarily, our perform first saves the matplotlib determine to a reminiscence buffer. It then encodes the binary PNG information as base64 textual content and wraps it within the desired information URL format.
Assuming we now have entry to the artificial every day order information, we will use the next perform to generate the plot and convert it into a correct information URL format in a single go:
def create_fig(df):
"""
Create a Matplotlib determine and convert it to a base64 information URL.
Weekends (Sat–Solar) are shaded.
Args:
-----
df: dataframe comprises one profile of every day order time collection.
dataframe has "date" and "orders" columns.
Returns:
--------
image_url: The information URL representing the determine.
"""
df = df.copy()
df['date'] = pd.to_datetime(df['date'])
fig, ax = plt.subplots(figsize=(8, 4.5))
ax.plot(df["date"], df["orders"], linewidth=2)
ax.set_xlabel('Date', fontsize=14)
ax.set_ylabel('Each day Orders', fontsize=14)
# Weekend shading
begin = df["date"].min().normalize()
finish = df["date"].max().normalize()
cur = begin
whereas cur <= finish:
if cur.weekday() == 5: # Saturday 00:00
span_start = cur # Sat 00:00
span_end = cur + pd.Timedelta(days=1) # Mon 00:00
ax.axvspan(span_start, span_end, alpha=0.12, zorder=0)
cur += pd.Timedelta(days=2) # skip Sunday
else:
cur += pd.Timedelta(days=1)
# Title
title = f'Each day Orders: {df["date"].min():%b %d, %Y} - {df["date"].max():%b %d, %Y}'
ax.set_title(title, fontsize=16)
# Format x-axis dates
ax.xaxis.set_major_formatter(mdates.DateFormatter('%b %d'))
ax.xaxis.set_major_locator(mdates.WeekdayLocator(interval=1))
plt.tight_layout()
# Acquire url
image_url = fig_to_data_url(fig)
return image_urlFigures 1-3 are generated by the above plotting routine.
2.4 Structured output
On this part, let’s talk about how to make sure the o3 mannequin outputs a constant JSON format as an alternative of free-form textual content. That is what’s often known as “structured output,” and it’s one of many key enablers for integrating LLMs into current computerized workflows.
To attain that, we begin by defining the schema that governs the anticipated output construction. We’ll be utilizing a Pydantic mannequin:
from pydantic import BaseModel, Area
from typing import Literal
from datetime import date
AnomalyKind = Literal["spike", "drop", "level_shift", "seasonal_outlier"]
class DateWindow(BaseModel):
begin: date = Area(description="Earliest believable date the anomaly begins (ISO YYYY-MM-DD)")
finish: date = Area(description="Newest believable date the anomaly ends, inclusive (ISO YYYY-MM-DD)")
class AnomalyReport(BaseModel):
when: DateWindow = Area(
description=(
"Minimal window that comprises the anomaly. "
"For single-point anomalies, use the interval that covers studying uncertainty, if the tick labels are unclear"
)
)
y: int = Area(description="Approx worth on the anomaly’s most consultant day (peak/lowest), rounded")
form: AnomalyKind = Area(description="The kind of the anomaly")
why: str = Area(description="One-sentence motive for why this window is uncommon")
date_confidence: Literal["low","medium","high"] = Area(
default="medium", description="Confidence that the window localization is right"
)Our Pydantic schema tries to seize each the quantitative and qualitative elements of the detected anomalies. For every subject, we specify its information sort (e.g., int for numerical values, Literal for a set set of selections, and so forth.).
Additionally, we use Area perform to supply detailed descriptions of every key. These descriptions are particularly vital as they successfully function inline directions for o3, in order that it understands the semantic which means of every element.
Now, we now have coated the multimodal enter and structured output, time to place them collectively in a single LLM name.
2.5 o3 mannequin invocation
To work together with o3 utilizing multimodal enter and structured output, we use LLM_client.beta.chat.completions.parse() API. A few of the key arguments embody:
mannequin: the deployment title;messages: the message object despatched to o3 mannequin;max_completion_token: the utmost variety of tokens the mannequin can generate in its closing response. Notice that for reasoning fashions like o3, they may generate reasoning_tokens internally to “suppose via” the issue. The presentmax_completion_tokensolely limits the seen output tokens that customers obtain;response_format: the Pydantic mannequin that defines the anticipated JSON schema construction;reasoning_effort: a management knob that dictates how a lot computational effort o3 ought to use for reasoning. The out there choices embody low, medium, and excessive.
We will outline a helper perform to work together with the o3 mannequin:
def anomaly_detection(instruction, fig_path,
response_format, immediate=None,
deployment="o3", reasoning_effort="excessive"):
# Compose messages
messages=[
{ "role": "system", "content": instruction},
{ "role": "user", "content": [
{
"type": "image_url",
"image_url": {
"url": fig_path,
"detail": "high"
}
},
]}
]
# Add immediate whether it is given
if immediate shouldn't be None:
messages[1]["content"].append({"sort": "textual content", "textual content": immediate})
# Invoke LLM API
response = LLM_client.beta.chat.completions.parse(
mannequin=deployment,
messages=messages,
max_completion_tokens=4000,
reasoning_effort=reasoning_effort,
response_format=response_format
)
return response.selections[0].message.parsed.model_dump()Notice that the messages object accepts each textual content and picture content material. Since we’ll solely use figures to immediate the mannequin, the textual content immediate is optionally available.
We set the "element": "excessive" to allow high-resolution picture processing. For our present case examine, that is almost certainly obligatory as we want o3 to raised learn high quality particulars like axis tick labels, information level values, and refined visible patterns. Nevertheless, keep in mind that high-detail processing would incur extra tokens and better API prices.
Lastly, through the use of .parsed.model_dump(), we flip the JSON output right into a normal Python dictionary.
That’s it for the implementation. Let’s see some outcomes subsequent.
3. Outcomes
On this part, we’ll enter the beforehand generated figures into the o3 mannequin and ask it to determine potential anomalies.
3.1 Spike anomaly
# df_spike_anomaly is the dataframe of the primary set of artificial information (Determine 1)
spike_anomaly_url = create_fig(df_spike_anomaly)
# Anomaly detection
end result = anomaly_detection(instruction,
spike_anomaly_url,
response_format=AnomalyReport,
reasoning_effort="medium")
print(end result)Within the name above, the spike_anomaly_url is the information URL for Determine 1. The output of the result’s proven beneath:
{
'when': {'begin': datetime.date(2025, 8, 19), 'finish': datetime.date(2025, 8, 21)},
'y': 166,
'form': 'spike',
'why': 'Single day orders leap to ~166, far above adjoining days that sit close to 120–130.',
'date_confidence': 'medium'
}We see that o3 mannequin faithfully returned the output precisely within the format we designed. Now, we will seize this end result and generate a visualization programmatically:
# Create picture
fig, ax = plt.subplots(figsize=(8, 4.5))
df_spike_anomaly['date'] = pd.to_datetime(df_spike_anomaly['date'])
ax.plot(df_spike_anomaly["date"], df_spike_anomaly["orders"], linewidth=2)
ax.set_xlabel('Date', fontsize=14)
ax.set_ylabel('Each day Orders', fontsize=14)
# Format x-axis dates
ax.xaxis.set_major_formatter(mdates.DateFormatter('%b %d'))
ax.xaxis.set_major_locator(mdates.WeekdayLocator(interval=1))
# Add anomaly overlay
start_date = pd.to_datetime(end result['when']['start'])
end_date = pd.to_datetime(end result['when']['end'])
# Add shaded area
ax.axvspan(start_date, end_date, alpha=0.3, coloration='crimson', label=f"Anomaly ({end result['kind']})")
# Add textual content annotation
mid_date = start_date + (end_date - start_date) / 2 # Center of anomaly window
ax.annotate(
end result['why'],
xy=(mid_date, end result['y']),
xytext=(10, 20), # Offset from the purpose
textcoords='offset factors',
bbox=dict(boxstyle='spherical,pad=0.5', fc='yellow', alpha=0.7),
arrowprops=dict(arrowstyle='->', connectionstyle='arc3,rad=0.1'),
fontsize=10,
wrap=True
)
# Add legend
ax.legend()
plt.xticks(rotation=0)
plt.tight_layout()The generated visualization seems to be like this:
We will see that the o3 mannequin accurately recognized the spike anomaly introduced on this first set of artificial information.
Not dangerous, particularly contemplating the truth that we didn’t do any standard mannequin coaching, simply by prompting an LLM.
3.2 Degree shift anomaly
# df_level_shift_anomaly is the dataframe of the 2nd set of artificial information (Determine 2)
level_shift_anomaly_url = create_fig(df_level_shift_anomaly)
# Anomaly detection
end result = anomaly_detection(instruction,
level_shift_anomaly_url,
response_format=AnomalyReport,
reasoning_effort="medium")
print(end result)The output of the result’s proven beneath:
{
'when': {'begin': datetime.date(2025, 8, 26), 'finish': datetime.date(2025, 9, 2)},
'y': 150,
'form': 'level_shift',
'why': 'Orders instantly leap from the 120-135 vary to ~150 on Aug 26 and stay elevated for all subsequent days, indicating a sustained baseline change.',
'date_confidence': 'excessive'
}Once more, we see that the mannequin precisely recognized {that a} “level_shift” anomaly is current within the plot:
3.3 Seasonality anomaly
# df_seasonality_anomaly is the dataframe of the third set of artificial information (Determine 3)
seasonality_anomaly_url = create_fig(df_seasonality_anomaly)
# Anomaly detection
end result = anomaly_detection(instruction,
seasonality_anomaly_url,
response_format=AnomalyReport,
reasoning_effort="medium")
print(end result)The output of the result’s proven beneath:
{
'when': {'begin': datetime.date(2025, 8, 23), 'finish': datetime.date(2025, 8, 24)},
'y': 132,
'form': 'seasonal_outlier',
'why': 'Weekend of Aug 23-24 exhibits order volumes (~130+) on par with surrounding weekdays, whereas different weekends constantly drop to ~115, making it an out-of-season spike.',
'date_confidence': 'excessive'
}It is a difficult case. However, our o3 mannequin managed to deal with it correctly, with correct localization and a transparent reasoning hint. Fairly spectacular:
4. Abstract
Congratulations! We’ve efficiently constructed an anomaly detection resolution for time-series information that labored totally via visualization and prompting.
By feeding every day order plots into the o3 reasoning mannequin and constraining its output to a JSON schema, the LLM managed to determine three totally different anomaly sorts with correct localization. All of this was achieved with out coaching any ML mannequin. Spectacular!
If we take a step again, we will see that the answer we constructed illustrates the broader sample of mixing three capabilities:
- See: multimodal enter to let the mannequin eat figures instantly.
- Suppose: step-by-step reasoning functionality to deal with complicated issues.
- Combine: structured output that downstream programs can simply eat (e.g., producing visualizations).
The mixture of multimodal enter + reasoning + structured output actually creates a flexible basis for helpful LLM purposes.
You now have the constructing blocks prepared. What do you wish to construct subsequent?
