Timofei Chernov, Fullstack developer
Accelerating Journalistic Research with AUTOPILOT's Automation
It seems AI is now a central part of our lives—where the tasks we assign it extend far beyond mere Q&A sessions. Consider the journalist’s flow: a powerful demonstration of how AUTOPILOT can transform a seemingly overwhelming research process into a seamless, efficient operation. Picture this: a journalist tasked with preparing for a critical interview must sift through data from a dozen websites, compile fascinating facts, craft pointed questions, and even generate answers to catch any discrepancies in real-time. What might have taken hours, even days, is accomplished in just a few minutes with AUTOPILOT.

But remember—this journalist’s flow is only one drop in an ocean of possibilities. AUTOPILOT’s flexibility is virtually limitless. Today’s demonstration serves as a prime example, yet its true potential extends to countless other applications. Whether you’re streamlining customer service, automating complex data analyses, or tailoring workflows to unique business needs, AUTOPILOT adapts to your creative demands with ease.

In just half an hour, you can set up an automation mechanism available to every member of your organization, delivering unmatched efficiency and quality. And while the journalist’s case illuminates one remarkable use, it’s merely the tip of the iceberg in a world of endless automation possibilities.
Quick Bite: If only your morning coffee could brew itself while AUTOPILOT does the heavy lifting!
The AUTOPILOT Workflow: A Step-by-Step Guide
Creating this flow is like setting up a mini assembly line for data. Here’s how you can build each component with precision:

1. Defining the Trigger

What Happens:
The flow kicks off with a simple input trigger. On the Dashboard, there's an input field where the journalist enters the subject’s name or company name.

Detailed Steps:
- Label Your Input: Give the trigger a clear name — for example, “Interviewed person (company)”
- Test the Input: Enter a sample name to simulate real data. Confirm that the system recognizes and processes the input correctly.
View in Autopilot
View on the Dashboard
2. Information Retrieval – Phase One: Recent Online Activity

What Happens:
At this step, we retrieve data reflecting recent online activity during the past 1-2 years.

Detailed Steps:
- Model Selection: We use a self-hosted model to ensure data privacy. This step avoids sending queries to external services like ChatGPT.
- Crafting the Prompt: Remember - better prompt - better result. It should capture recent activity and return an array of source data, including websites and timestamps.
- Injecting the Trigger Value: Insert the value obtained from the trigger into the prompt to personalize the query.
- Testing: Run a test to ensure the results include website arrays and data snapshots.
All prompts improved by AI
3. Information Retrieval – Phase Two: General Background Information

What Happens:
Next, we switch focus to general information about the subject, using a distinct prompt and model that isn’t reliant on live internet searches.

Detailed Steps:
- Model Selection: Choosing a secure, self-hosted model.
- Custom Prompt: Input a different prompt designed to fetch background details, integrating the trigger value to tailor the output.
- Testing: Verifying that the response contains broad details about the subject that are useful for contextual understanding.
4. Synthesizing Data into a Readable Article

What Happens:
With both sets of data in hand, we now combine them into a structured article.

Detailed Steps:
- Utilizing Mistral: Selecting the Mistral model for its capability to transform and organize text.
- Consolidating Responses: Insert outputs from the two previous steps into a pre-designed prompt that instructs Mistral to merge the data into an article format.
- Testing and Refinement: Executing the step and reviewing the preliminary article. Fine-tune the prompt if necessary to enhance clarity or coherence.
5. Extracting Key Facts

What Happens:
The flow then distills the synthesized information, pinpointing the most compelling facts about the subject.

Detailed Steps:
- New Step Setup: Adding a step that identifies unique interesting facts and facets.
- Prompt Detailing: Using a prompt that instructs the model to extract and highlight key facts—almost like the “highlight reel” of the subject's background.
- Testing: Verifying that the output correctly lists essential facts without overwhelming the reader with trivial details.
6. Generating Structured Interview Questions

What Happens:
Leverage AI to transform the article data into well-structured, prioritized interview questions.

Detailed Steps:
- Model Choice – DeepSeek: Selecting DeepSeek for its precision in generating targeted queries.
- Prompt Integration: We input a prompt along with the previously gathered data to instruct DeepSeek to create questions, ensuring each one is marked by its importance.
- Step Testing: Running the step and inspecting the list of categorized questions, ensuring they are sharp and relevant.
7. Crafting Answer Attempts

What Happens:
After question generation, the flow further attempts to provide preliminary answers using available data or additional internet access.

Detailed Steps:
- Internet-Enabled Model: Opt for a model with live internet capabilities to source additional information, or rely on the already gathered data.
- Answer Prompt: Using a detailed prompt that instructs the model to answer each question, either retrieved from prior research or logically deduced.
- Test and Validate: Running a test to ensure answers are coherent, and factual, or provide qualified assumptions to guide further research.
8. Displaying Final Results on the Dashboard

What Happens:
The concluding step is to consolidate and display all outputs on a unified dashboard view.

Detailed Steps:
- Integrating a Plugin: Using a "Wize Verify Content" plugin to assign clear labels for each piece of output—facts, questions, answers, and the composed article.
- Field Organization: Create a field for each result and map the data values.
- Final Testing: Executing the full flow from the Dashboard. Clicking on an appeared execution will expand the details, showcasing every sequential step.
Results
All of the results can be formatted
Conclusion: Streamlined Research Redefined

By configuring this flow in just 30 minutes, AUTOPILOT transitions routine journalistic tasks into an efficient process. All it takes is entering a name, and the system delivers a comprehensive package that includes background articles, insightful facts, structured interview questions, and even preliminary answers—all executed in under 5 minutes.

For journalists who’ve long wished for a way to “fast-forward” the mundane parts of research (and let’s be honest, who hasn’t dreamed of that extra coffee break?), AUTOPILOT offers the perfect solution. In our upcoming sessions, we’ll explore even new functionalities and integration capabilities, so stay tuned for more ways to reinvent your workflow.

Remember, in the world of fast-paced media, when AUTOPILOT is at the helm, even the busiest days become just a matter of a few clicks!

Now, gear up to pilot your research with AUTOPILOT—because if your research could drive, wouldn’t you want it to be on cruise control?
20 February, 2025