How to Turn Data into Compelling Stories with LLMs

Introduction: the critical question

How can you use Large Language Models (LLMs) to transform raw data and incomprehensible numbers into engaging, easy-to-understand stories? The short answer is: with a well-structured prompt and the right tools, you can automate the entire data storytelling process, from data cleaning to the final narrative.

Why data storytelling is essential in 2026

Today, companies generate more data than ever, but statistical analysis alone is not enough. A good data story: - Accelerates decision-making - Makes insights understandable to everyone, even non-technical people - Increases engagement and memorability of reports Modern LLMs, such as Nemotron 3.5 Transcribe and Dessa, offer unique capabilities that simplify the creation of dynamic and multimodal narratives.

The new LLMs that are changing the game

• Nemotron 3.5 Transcribe
• Dessa
• Hermit

How to build effective prompts for data storytelling

The key is not the complexity of the model, but the clarity of the prompt. Follow this structure:

1. Define the goal: What decision will the audience need to make?
2. Identify the audience: Technical level, industry, preferences.
3. Choose the format: Text, slides, audio, interactive visualization.
4. Provide the data: Attach CSV, JSON, or a URL.
5. Request the structure: Introduction, progression, anomalies, forecasts, call-to-action.

Example prompt(suitable for Nemotron or any text LLM):Analyze the following sales dataset (CSV attached) and produce a 3-act story: 1. Introduction: overall trend and market context. 2. Progression: month by month, highlighting peaks and declines. 3. Future outlook: 3 insights and suggestions for the next quarter. Use a professional tone, include markdown tables, and suggest a possible slide for the presentation. Dataset:

Practical example: from a CSV file to a story in 5 steps

Upload the CSV file to a Jupyter environment. Use pandas for an initial overview.import pandas as pd import numpy as np df = pd.read_csv('sales_2026.csv') print(df.head()) print(df.info())

If you have recorded customer interviews, convert the audio to text using Nemotron 3.5 Transcribe.from transformers import pipeline asr = pipeline('automatic-speech-recognition', model='NVIDIA/Nemotron-3.5-Transcribe') result = asr('customer_feedback.wav') text = result['text'] # Save to a JSON for the next step import json json.dump({'transcript': text}, open('feedback.json', 'w'))

Use an LLM (e.g., an open-source model hosted on Hugging Face) to create the story.from transformers import AutoModelForCausalLM, AutoTokenizer tokenizer = AutoTokenizer.from_pretrained('meta-llama/Llama-2-7b-chat-hf') model = AutoModelForCausalLM.from_pretrained('meta-llama/Llama-2-7b-chat-hf') def generate_story(prompt): inputs = tokenizer(prompt, return_tensors='pt') outputs = model.generate(**inputs, max_length=800, num_return_sequences=1) return tokenizer.decode(outputs[0], skip_special_tokens=True) prompt = """Analyze the attached sales dataset and produce a 3-act story. Dataset: """ + open('sales_2026.csv').read() story = generate_story(prompt) print(story)

Transform the story into audio using Dessa for an engaging narrative.import requests def text_to_speech_dessa(text, voice='professional'): payload = {'text': text, 'voice': voice} resp = requests.post('https://api.block.ai/dessa/speech', json=payload) return resp.content audio_bytes = text_to_speech_dessa(story) with open('story.mp3', 'wb') as f: f.write(audio_bytes)

Hermit allows you to create an agent that can answer questions about the dataset, filter insights, and even regenerate the story for different target audiences.# Example of configuring a Hermit agent from hermit import HarnessAgent agent = HarnessAgent( name='StoryAgent', tools=['data_loader', 'analyzer', 'generator'], memory=['sales_2026.csv', 'feedback.json'] ) question = 'Which products have the highest ROI in quarter 3?' answer = agent.query(question) print(answer)

Current trends: integrating multimodal LLMs into workflows

Companies are moving from static pipelines to dynamic storytelling loops:

• Audio-based pipelines: use Nemotron to transcribe meetings, Dessa to generate vocal summaries.
• Data agents: Hermit orchestrates complex workloads, linking datasets, models, and user interfaces into a single story.
• Engineered prompts: the industry is adopting structured prompt templates (like the 5-step model above) to standardize story quality.

Integration tips

• Start with a small prototype: load a dataset, generate a story, and present it to your team.
• Use open-source adapters (Nemotron, Dessa, Hermit) to maintain control over your data.
• Document every prompt: technical and non-technical teams can then replicate and improve results.

Common mistakes and how to avoid them

• Prompt overloading: an overly long prompt can generate irrelevant responses. Break it down into sub-tasks.
• Poor quality data: LLMs reflect the quality of their training data. Cleaning and validation are essential.
• Lack of context: not specifying the audience or objective can lead to generic narratives. Always include context and tone.

Conclusion: transform data into powerful stories

Next steps:

1. Choose a small dataset (CSVs or audio files) and test the workflow described above.
2. Create a prompt template that fits your industry and share it with your team.
3. Explore Dessa and Hermit's features to add voice and interactivity to your stories.

The future of data storytelling is now, and LLMs are its engine. Start telling data-driven stories faster, clearer, and more memorably.

IMPORTANT: We are in 2026. All temporal references must be current. Do not mention past years such as 2024, 2023, etc. The content must be fresh and relevant to the present. CURRENT CONTEXT (October 2026): Here are some recent trends and news stories for inspiration: - Infini-AI-Lab Releases Hermit, an Open-Source Harness Adapter Framework for AI Agents: Infini-AI-Lab released Hermit, an Apache-2.0 harness adapter framework that trains an external layer of instructions, skills, memory and tools from ta... [2026-10-29] - Google DeepMind Introduces Gemini 4 Carbon, Its Top Model for Cyber Defense and Coding: Google DeepMind introduced Gemini 4 Carbon, a frontier model that scores 59 on the Artificial Analysis Intelligence Index and 88.2% on SEC-bench Pro. ... [2026-10-28] - Block Unveils Dessa, a Full-Duplex Speech Model Trained Only on Speech: Block’s AI lab released Dessa, a full-duplex speech-to-speech model trained only on human conversation audio. It runs locally on a MacBook Pro M4 Ma... [2026-10-29] Use this current information as inspiration to create an original and relevant prompt for 2026.