How to Turn Raw Data into Stories with Open-Weight LLMs

Introduction: Why LLMs are essential for data storytelling

Why LLMs are perfect for data storytelling

Contextual understanding:LLMs automatically analyze complex datasets, identify trends, and detect anomalies.
Natural language generation:They create titles, paragraphs, and key points fluidly, tailored to the audience.
Customization:They can be guided by industry-specific prompts to adapt tone, voice, and detail level.
Integration with modern APIs:Services likeYou.com Inference Routeroffer OpenAI-compatible API access to 12 open-weight models with 99.9% uptime.
On-device TTS:Models likePhonon from Gradium(31M parameters, 64.5 MB) enable voice narration generation directly on devices.

The new open-weight tools of 2026: real-world examples

You.com Inference Router

Launched in October 2026, Inference Router provides a unified API endpoint for 12 open-weight models. Key features include:

99.9% SLA uptime and exclusive hosting in the United States.
Zero data retention for enhanced privacy.
OpenAI compatibility for seamless integration.

These advantages make Inference Router ideal for data storytelling pipelines requiring reliability and data control.

Phonon from Gradium: On-device TTS

Phonon is an open-source (Apache 2.0) text-to-speech model family that occupies only 64.5 MB. Its small size enables voice narration generation on edge devices, perfect for data-driven dashboards, reports, or podcasts.

TinyFish's Agentic Search

TinyFish has published a two-article guide defining a7-phase search architecturefor production AI agents. This architecture includes trial contracts and evidence, allowing LLMs to autonomously retrieve data, prepare datasets, and provide context for story generation.

Building a data storytelling workflow with LLMs

Below is a practical workflow that combines the latest services.

1. Data ingestion and summarization

Start by uploading the raw dataset (CSVs, databases, APIs). Use an LLM via Inference Router to generate a concise summary:

import requests
import json

API_URL = "https://inference.you.com/v1/chat/completions"
HEADERS = {"Authorization": "Bearer YOUR_KEY"}

def summarize_data(data_sample):
    prompt = f"""
    Provide a concise summary of up to 150 words for the following data sample:
    {data_sample}
    Highlight trends, key values, and any anomalies.
    """
    payload = {
        "model": "youcom-inference-router",
        "messages": [{"role": "user", "content": prompt}]
    }
    resp = requests.post(API_URL, headers=HEADERS, json=payload)
    return resp.json()["choices"][0]["message"]["content"]

2. Creating a narrative

Use a structured prompt to obtain a complete story:

def generate_story(summary):
    prompt = f"""
    You are a data storytelling expert. Write a story in Italian that includes:
    1. An engaging title (max 10 words)
    2. Key points with numbers and percentage changes
    3. A narrative paragraph explaining the trends
    4. A bulleted list of recommendations

    Data summary:
    {summary}
    """
    payload = {
        "model": "youcom-inference-router",
        "messages": [{"role": "user", "content": prompt}]
    }
    resp = requests.post(API_URL, headers=HEADERS, json=payload)
    return resp.json()["choices"][0]["message"]["content"]

3. Voice narration with Phonon

Once you have the story, convert it to audio using the local Phonon model:

from transformers import pipeline

tts = pipeline("text-to-speech", model="gradium/phonon-31m")
audio = tts("The generated story above: ...")  # insert the full story text

4. Integrating agent-based search

Incorporate TinyFish's search steps to retrieve up-to-date data (e.g., new store locations, market trends) before generating the story:

def fetch_live_data(query):
    # Example using TinyFish's search architecture
    agent = TinyFishAgent(phases=7, evidence_contracts=True)
    return agent.search(query)

Practical example: Turning monthly sales data into a story

Imagine a dataset recording monthly sales by region:

January: North $120k (+5%)
February: South $98k (-2%)
March: Central $135k (+12%)

sample = "{January: North $120k (+5%), February: South $98k (-2%), March: Central $135k (+12%)}"
print(summarize_data(sample))

summary = summarize_data(sample)
story = generate_story(summary)
print(story)

tts("Here's your story: " + story)The result is a shareable dashboard, PDF, or podcast that automatically tells the story of sales performance.

Common mistakes and how to avoid them

Sending raw datasets to the LLM:Limit input to a summary or key statistics to reduce token costs and improve clarity.
Generic prompts:Use structured prompts (title, key points, narrative, recommendations) to get consistent results.
Neglecting validation:Always verify factual claims with reliable sources or TinyFish's trial contracts.
Ignoring privacy regulations:Leverage Inference Router's zero data retention benefits to comply with GDPR and CCPA regulations.

Takeaway: Put data storytelling with LLMs into practice today

Sign up for You.com Inference Routerand get immediate access to reliable open-weight models.
Integrate Gradium's Phononto add voice narration to your reports.
Design an agent-based search workflow(inspired by TinyFish) to automatically update datasets.
Experiment with our prompt templateto quickly create consistent stories.

Conclusion

In 2026, LLMs combined with new open-weight models like Inference Router, Phonon, and TinyFish's search architecture make data storytelling faster, more secure, and richer in nuance than ever before. Whether you're turning a spreadsheet into a podcast or powering a dashboard with real-time narrative updates, these techniques provide a practical, production-ready edge. Start today, experiment with prompts, and watch your data come alive in stories your audience can see, read, and hear. IMPORTANT: We are in 2026. All temporal references must be current. Do not mention past years like 2024, 2023, etc. The content must be fresh and relevant to the present. CURRENT CONTEXT (October 2026): Here are some recent trends and news to inspire you: - Gradium Open-Sources Phonon, a 31M-Parameter On-Device TTS Model Under Apache 2.0: Gradium has open sourced Phonon, a family of on-device text-to-speech models. The English Phonon 31m ships under Apache 2.0, fits in 64.5 MB, and star... [2026-10-22] - Google Cloud Launches Agent Compute Engine: A MicroVM per AI Agent: Google Cloud announced Agent Compute Engine, a compute service that gives every AI agent its own hardware-isolated microVM with no credentials inside.... [2026-10-23] - TinyFish Breaks Down Agentic Search Architecture for Production AI Agents: TinyFish published two engineering research articles on agentic search. One defines a 7-phase production architecture with evidence contracts and boun... [2026-10-23] Use this current information as inspiration to create an original and relevant prompt for 2026.