Using the Cogynt AI Chatbot
The Cogynt AI Chatbot is accessed within Workstation, and is designed to help analysts efficiently analyze events for insights. With proper chat configuration, the bot can summarize events, identify people, highlight data quality issues, advise on deeper areas for analysis, and more.
Cogynt's AI Chatbot is a predictive tool, and quality of results can vary. As with any AI interaction, all results should be verified by a human.
This page assumes that users have the Intelligence: Chatbot: View & Edit permissions, and that an AI Config with a system prompt has been assigned to an event type.
Creating Sessions
To create sessions with the AI chatbot, first open an event in Object Details Viewer that the bot can interact with.
- From any event open within Object Details Viewer, or its own tab, at the top right side of the event, click the Cogynt AI icon. The AI chat session opens.
- Interact with the chatbot:
- At the bottom of the chatbot modal, enter any question or select a suggested question.
- Click the More menu (⋮) and select New Chat to create a new session with the AI chatbot.
- At the top right side, click the Chat History dropdown to locate previous sessions for review. Click the user's name to review a transcript of the chat history.
- At the top right side of the chatbot modal, shrink or expand the size of the modal window.
- At the top right side, click the X icon to close the AI chatbot modal.
Whenever an event's system prompt receives an edit or update, create a new session to be sure you're using the most up-to-date version of that prompt.
Using the AI Chatbot
To use the chatbot, type questions into it. Generally, questions focused on a single thread of thought are best. Suggested Questions are immediately visible whenever a new session is opened. At the bottom of the chat window, click a prompt and the bot will produce results aimed at answering the prompt.
The following usage tips are aimed at improving the quality of the results your team gets from the bot.
Validating Chatbot Results
The chatbot, like all AI applications, is influenced by the prompt it is given. Coupled with its own internal logic and possible data issues, it is important for users to carefully consider and review all chatbot results for accuracy. If accuracy is off, you can improve system prompts, review your suggested questions, or try to rephrase your query.
The less specific a question is, the longer the bot will take to respond and the lower the chance for a succinct and accurate response.
Improving System Prompts
Before beginning a new session with the AI Chatbot, we suggest that users understand the existing system prompt for an event type.
Never consider the chatbot's system prompt to be a "set and forget" solution. As your organization grows, it brings new ideas and methodologies that the bot should also utilize. Adjusting system prompts to introduce these new ideas will change the approach and the results to expect from the chatbot.
Improving on Suggested Questions
Suggested Questions are meant to efficiently access information your team routinely needs from specific event types. Good suggested questions prompt users and get them to actionable data. As you experiment with the bot, you may uncover new ideas for suggested questions.
Using New Sessions to Separate Ideas
Use a new chat session to query an event from a different angle of analysis. This both separates each thread of thinking for review later, and offers the chatbot a more succinct chat session to improve reliability and focus on topic accuracy.
As an example, an event about a crime might inspire questions about who was involved, and what was committed. One session could focus on the who, the other session on analyzing the scene of the crime.
Reviewing Chatbot History
Reviewing the chatbot's history is helpful for providing context for the bot's analysis. It is beneficial to review the history not only from your user, but other users as well. You may uncover new insights, or a different approach to your data that inspires further analysis.