Summarize Contents

With Graphlit, you can use LLMs to summarize contents. We provide a set of default summarization options, such as bullet points, social media posts, and followup questions, or you can provide a custom summarization prompt.

By default, we use the Azure OpenAI GPT 3.5 Turbo 16k model for summarization, but you can select your own custom specification to use other models.

You can summarize a specific content item, by specifying the ID, or you can filter all of your content, and summarize multiple contents at once.

For more information about content filtering arguments, see here.

Summarize Content

g summarize --type content --id "{id}"

[
  {
    "content": {
      "id": "6d80b989-5e94-4bfb-83db-a7ea82fdf640"
    },
    "type": "SUMMARY",
    "items": [
      {
        "text": "In this episode of the Mapscaping podcast, host Daniel talks with Kirk Marple, founder and CEO of Unstruct Data. They discuss the concept of unstructured data, which encompasses a broad set of file-based data such as imagery, audio, 3D geometry, point clouds, documents, and email. Kirk explains that while there is always a structure to these file types, the term \"unstructured data\" is used to differentiate from the structured modern data stack world. He also introduces the concept of first order, second order, and third order metadata, which represent the increasing levels of context and inference that can be derived from unstructured data.\n\nKirk and Daniel explore the potential of edge computing in relation to unstructured data, discussing how pushing compute closer to the source of data can impact the generation and management of metadata. They also delve into the idea of creating a network effect for data by building knowledge graphs that connect and contextualize data from various sources. Kirk shares examples of projects where Unstruct Data has helped companies extract insights from their unstructured data, such as an aerial survey company looking to improve their data management and search capabilities.\n\nThe conversation touches on the challenges of managing the growth and enrichment of knowledge graphs, as well as the importance of focusing on specific use cases and industries to avoid becoming overwhelmed. Kirk explains that Unstruct Data primarily focuses on unstructured data that is perceiving real-world assets, with a geospatial element. He envisions a future where their technology can provide a contextual \"heads up display\" for physical assets by linking data from various sources, such as photos, sensor data, and databases.\n\nKirk emphasizes that while their approach shares similarities with companies like Palantir, Unstruct Data aims to provide a more accessible, no-code environment for businesses of various sizes to build upon their data platform. He sees potential for vertical applications to be built on top of their technology, such as property inspection solutions that integrate photos, emails, and other relevant data. The episode concludes with Kirk inviting listeners to connect with him on LinkedIn and visit the Unstruct Data website to learn more about their offerings and discuss potential use cases.",
        "tokens": 432,
        "summarizationTime": "PT42.312699S"
      }
    ]
  },
  {
    "content": {
      "id": "6d80b989-5e94-4bfb-83db-a7ea82fdf640"
    },
    "type": "QUESTIONS",
    "items": [
      {
        "text": "What is unstructured data and how does it differ from structured data?",
        "tokens": 14,
        "summarizationTime": "PT20.9607848S"
      },
      {
        "text": "How can metadata be used to provide context and meaning to unstructured data?",
        "tokens": 15,
        "summarizationTime": "PT20.9607848S"
      },
      {
        "text": "What are some examples of how computer vision and machine learning can be applied to extract insights from unstructured data like images and video?",
        "tokens": 26,
        "summarizationTime": "PT20.9607848S"
      },
      {
        "text": "How can knowledge graphs be leveraged to connect entities and relationships extracted from unstructured data?",
        "tokens": 18,
        "summarizationTime": "PT20.9607848S"
      },
      {
        "text": "What are some potential enterprise applications and industry use cases for an unstructured data platform?",
        "tokens": 17,
        "summarizationTime": "PT20.9607848S"
      }
    ]
  }
]

Custom Summary

In addition to the built-in summarization options, you can provide a custom summarization prompt.

Example:

Identify the main topics and interesting points in the text. Write detailed, verbose bullet points for each. Be specific when referencing persons, organizations, places, or any other named entities.  Make sure to identify places and addresses, and extract and deduplicate any contact information for persons or organizations mentioned.

Summarize Audio Files

With content filtering, you can summarize a subset of your content, such as audio files.

For podcasts and YouTube videos, Graphlit offers automatic chapter summarization.