Navigation
‣ Home
Data sharing & project security
This is a list of things we are tinkering with or know that we will eventually want to build.
Updated Nov 20, 2024
Currently, Tato reads from project management apps and listens to calls. Teams often discuss tasks outside these apps, especially during scrums and meetings. Manually updating task statuses afterward is time-consuming. We aim to streamline this process by enabling Tato to write back to these apps.
Phase 1: Identify Status Mismatches complete
Tato detects discrepancies between discussed tasks and their statuses in project management apps, flagging them for review.
Phase 2: Comment Back to Tickets. upcoming
Tato will automatically add comments to tickets mentioned in discussions, providing updates based on meeting conversations.
Phase 3: Automate Status Updates future plan
Tato will change task statuses, toggle blocked flags, and reassign tasks directly within project management apps.
Ensuring Tato accurately interprets discussions to prevent erroneous updates. Maintaining user trust by providing transparent reasoning for any changes made.
Begin development of Phase 2, focusing on accurate comment generation. Collect user feedback from Phase 1 to improve mismatch detection.
Updated Nov 20, 2024
Currently, Tato enhances team communication by sending missed meeting summaries, daily or weekly project updates, and reminders to include Tato in meetings. We're thinking through how we may extend Tato's capabilities by integrating it directly into chat platforms like Microsoft Teams and Slack. How might people ask questions in the tools they already use? How might people give Tato updates in those same tools and workflows?
Phase 1: Interactive Q&A within Chat Tools future plan
Enable users to ask Tato questions about their projects directly within Teams or Slack.
Phase 2: Two-Way Communication via Chat and Voice future plan
Allow users to send updates back to Tato through chat messages, text messages, or voice calls, making interaction more flexible and convenient.
Prioritization between phases may shift based on customer needs and production timelines.
Experiment notes updated Nov 20, 2024
Visuals and KPIs can be an effective way of directing the users attention to something while giving them a visual of the overview so they can see the forest from the trees. Tato generates thousands of snippets, so we need to evaluate and quantify the data based on what needs attention to make use of the data that needs to be considered by project team members.
Our first experiment with traffic lights is to generate green for something that indicates a good fact that does not need further attention, red for something that requires attention, and yellow for something that is somewhere in between. The results of providing the traffic lights to users were bad at first because people did not know what had been used to come up with the color, so they did not trust the data. We then added a reason field where Tato explains why the snippet is set to a color, and users started to like the traffic lights.
Then, we faced the challenge of certain workflows having an overwhelming amount of snippets to review. We need to roll-up snippet traffic lights to components that they are related to. This sounds simple, but we have not yet found the right rules or evaluations to come up with traffic light roll-ups that are helpful to the customer.
Next, we will probably try showing underlying snippet stats with an overall rollup traffic light and some reasoning. We believe that the new reasoning models such as o1 will perform better than GPT4o, but we have not yet tested it.
Experiment notes updated Nov 20, 2024
Conversations go back and forth between different topics. Identifying a shift in topics, and summarizing the state of a topic is going well when we are summarizing a call, but snippet generation needs to go one step further if we want to make the most of the data. For example, knowing that a topic went back and forth between decisions 5 times during a call, and was discussed for 53 minutes is important information, because the data would tell a very different story if it was 1 simple decision made as a result of discussing the topic for 2 minutes.
Currently, snippets are extracted based on textual data, but not based on numerical data. We think that our extraction model might need to go deeper on the numerical data, and apply a hierarchical model to extracting topics, which would tell me that a topic was discussed from timestamp A to timestamp B, and then allow me to define sub-topics discussed within that timeframe. That would allow Tato to generate statistics such as quantity of time a topic was discussed over several calls, and degree to which alignment is high within the project stakeholders.
Experiment notes updated Nov 20, 2024
We believe that hierarchy trees are the best data model to represent scope. This is also well alignment to most project management tools and methodologies. In an ideal world, Tato would read documents such as the project charter, the hierarchies from the project management apps, and aggregate that with conversation data to determine the perfect representation of the scope/hierarchies. We are approaching this from a variety of different steps and experiments:
<aside> <img src="notion://custom_emoji/73b473c9-9304-4189-b3bb-9695ac1713e9/1436baa9-b63a-8058-aaf1-007a93141e6a" alt="notion://custom_emoji/73b473c9-9304-4189-b3bb-9695ac1713e9/1436baa9-b63a-8058-aaf1-007a93141e6a" width="40px" /> 👋 Reach out with any questions [email protected]
</aside>
Table of Contents