Adding AI to your apps and workflows

Artificial intelligence (AI) is increasingly becoming an integral component of modern apps and workflows. In this article, we'll explore how you can leverage AI, particularly large language models (LLMs), within the context of Retool applications. We'll take you step-by-step through a couple of demos showcasing practical implementations of AI to streamline tasks such as PDF parsing, data extraction, and spam detection in workflows.

Retool Community and State of AI Report

Before diving into the demos, it's worth highlighting the Retool community and the recently released State of AI report. The community page serves as a hub where you can find recordings of sessions and post any unanswered questions from Q&A segments. Meanwhile, the State of AI report, compiled from a survey of 750 individuals in the tech industry, gives valuable insights into how AI is perceived and used in organizations today.

Demo: Using LLMs for PDF Parsing in Retool

Step 1: Upload PDF

We start by creating a Retool app that allows users to upload PDFs. This can be particularly useful for roles like real estate agents dealing with extensive property descriptions. We begin by adding a file drop component labeled appropriately.

Step 2: Parse PDF

Once a PDF is uploaded, we control an AI action query to convert the document to text format for easier data manipulation. This involves selecting the "Convert Document to Text" action and linking it to our file drop zone.

Step 3: Generate a Listing Summary

Next, we create another AI action to summarize the parsed text. This query uses the "Generate Text" option to summarize key pieces of information relevant to a real estate listing, which is then displayed in the app's UI.

Step 4: Interactive Question and Answer

To further interact with the extracted data, we add a chat component. This chat allows users to ask detailed questions about the uploaded PDF, leveraging additional prompts to provide contextual answers based on the raw data.

Using Amazon Bedrock Models

You can extend this functionality by integrating Amazon Bedrock models, including options like GPT-4, Meta Llama 2, and others. Testing different models for the same queries can help determine the best fit for accuracy and cost-efficiency.

Demo: Workflow Automation and Spam Detection

Step 1: Basic Setup

For workflow automation, we create a contact form in Retool with data being stored in the Retool Database (Retool DB). The database tracks columns such as feedback and full names submitted through the form.

Step 2: Implement AI for Spam Detection

Using a workflow, we implement an AI action to parse feedback submissions and flag potential spam. The AI action is triggered upon form submission, processing the form data and updating the database to mark spam entries.

Step 3: Retrieving Data

To test, we submit various entries through the form and verify the workflow's effectiveness by checking the Retool database for accurate spam flagging based on previous AI analysis.

Vector Databases

Vector databases store textual and other data in a vectorized form, allowing for efficient similarity searches and contextual relevance determinations. Retool now supports vector databases, making it easier to integrate into AI workflows.

Conclusion

Retool provides a versatile platform for incorporating AI into apps and workflows. With native support for various LLMs and easy integrations, creating intelligent applications for tasks like document parsing and workflow automation becomes highly accessible.

If you have any questions or need further information, please visit the Retool Products page.

Keywords

Artificial Intelligence (AI)
Retool
Large Language Models (LLMs)
PDF Parsing
Workflow Automation
Amazon Bedrock
Vector Databases

FAQ

Q: How can I integrate custom assistants from the OpenAI API with Retool's chat component? A: You can use REST API calls to integrate custom assistants from OpenAI with the Retool chat component. You would need to set up these calls from within Retool.

Q: Can Retool handle large PDFs for parsing, such as documents 50 pages or more? A: Handling larger PDFs may require consideration of the AI model's context window limits. Utilizing Retool Vectors can help by providing only the most relevant pieces of context to the language model, thus managing larger document sizes more efficiently.

Q: Is it possible to compare multiple PDFs using Retool’s AI functionalities? A: Yes, you can upload multiple documents, and set up AI action queries to compare them. You can provide the AI with context from both documents to answer comparative questions.

Q: Are Retool’s AI functionalities multilingual? A: The Retool AI team is actively working on improving multilingual support within the platform. Stay tuned for updates in this area.

Q: What is the best way to integrate an external Vector database with Retool? A: While Retool Vectors provide native support, you can also integrate external Vector databases using their APIs via REST API queries in Retool.

By leveraging Retool's powerful AI features, you can streamline complex tasks, build smarter applications, and drive greater efficiencies in your workflows. Whether you're parsing intricate documents or automating spam detection, Retool makes it accessible and feasible.