Sentiment Analysis, Text Classification, and Named Entity Recognition with LLMs: A Practical Guide

#learnai #oxlo #ai

We are building a lightweight text intelligence pipeline that runs sentiment analysis, topic classification, and named entity recognition in a single LLM call. It is useful for support teams, analysts, and developers who need structured insights from unstructured text without maintaining three separate models. Because Oxlo.ai uses flat request-based pricing regardless of input length, you can feed in long support threads or documents without watching costs scale with tokens.

What you'll need

Python 3.10 or newer
An Oxlo.ai API key from https://portal.oxlo.ai
The OpenAI SDK: pip install openai

Step 1: Configure the Oxlo.ai client

First, import the SDK and point it at the Oxlo.ai endpoint. I use Llama 3.3 70B here because it follows mixed instructions and JSON mode reliably, but you can swap in Qwen 3 32B, Kimi K2.6, or DeepSeek V3.2 without changing anything else.

from openai import OpenAI

client = OpenAI(base_url="https://api.oxlo.ai/v1", api_key="YOUR_OXLO_API_KEY")

response = client.chat.completions.create(
    model="llama-3.3-70b",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Hello world"},
    ],
)
print(response.choices[0].message.content)

Step 2: Write the system prompt

I keep the prompt in a constant so I can tweak labels or categories without touching the Python logic. This instruction forces the model to return a single JSON object containing sentiment, category, and named entities.

SYSTEM_PROMPT = """You are a text analysis engine. Given a user message, output a single JSON object with exactly these keys:
- sentiment: one of "positive", "negative", "neutral", or "mixed"
- category: one of "support", "billing", "sales", or "general"
- entities: an array of objects, each with "text" and "label" (PERSON, ORG, PRODUCT, LOCATION)

Rules:
- Do not include markdown formatting.
- Output only valid JSON.
- If no entities are found, return an empty array."""

Step 3: Implement the analysis function

Now wrap the API call in a function. I enable JSON mode to keep the output structured. On Oxlo.ai, request-based pricing means the cost is the same whether you send a short sentence or a long support transcript, so this pattern stays predictable for bulk jobs.

import json
from openai import OpenAI

client = OpenAI(base_url="https://api.oxlo.ai/v1", api_key="YOUR_OXLO_API_KEY")

SYSTEM_PROMPT = """You are a text analysis engine. Given a user message, output a single JSON object with exactly these keys:
- sentiment: one of "positive", "negative", "neutral", or "mixed"
- category: one of "support", "billing", "sales", or "general"
- entities: an array of objects, each with "text" and "label" (PERSON, ORG, PRODUCT, LOCATION)

Rules:
- Do not include markdown formatting.
- Output only valid JSON.
- If no entities are found, return an empty array."""

def analyze_text(text: str) -> dict:
    response = client.chat.completions.create(
        model="llama-3.3-70b",
        messages=[
            {"role": "system", "content": SYSTEM_PROMPT},
            {"role": "user", "content": text},
        ],
        response_format={"type": "json_object"},
    )
    raw_output = response.choices[0].message.content
    return json.loads(raw_output)

Run it

Test the pipeline with a couple of real-world snippets. The first sample mixes praise and criticism, while the second is a plain negative support ticket.

if __name__ == "__main__":
    samples = [
        "Sarah from TechFlow said the new analytics suite is a game changer, though she finds the invoice portal confusing.",
        "I am furious that my shipment from Berlin was delayed again with no explanation.",
    ]

    for text in samples:
        result = analyze_text(text)
        print(json.dumps(result, indent=2))

Expected output looks like this:

{
  "sentiment": "mixed",
  "category": "billing",
  "entities": [
    {"text": "Sarah", "label": "PERSON"},
    {"text": "TechFlow", "label": "ORG"},
    {"text": "analytics suite", "label": "PRODUCT"},
    {"text": "invoice portal", "label": "PRODUCT"}
  ]
}
{
  "sentiment": "negative",
  "category": "support",
  "entities": [
    {"text": "Berlin", "label": "LOCATION"}
  ]
}

Wrap-up

You can drop this function into a FastAPI endpoint or a Celery worker to process tickets as they arrive. If you need deeper reasoning for ambiguous text, switch the model to kimi-k2.6 or deepseek-v3.2 on Oxlo.ai and leave the rest of the code unchanged.

DEV Community