SideTrack

Overview

SideTrack is an AI application that automatically visualizes your conversation as a diagram, real-time.

Process

At an AI hackathon, our team created 'Tangent,' a prototype that could identify when a speaker had gone off-topic. Tangent tied for 3rd place, and this early success encouraged us to further develop the concept of what is now ‘SideTrack.’

Competitive Review and Market Research

Before diving deep into SideTrack’s development, I conducted a competitive review to get a sense of where SideTrack might have product market fit, and what challenges might lie ahead. This research confirmed many of my assumptions, as well as illuminated some worthwhile discoveries:

• There are no other tools that provide the same value proposition or proposed combination of features (real-time transcription, summaries, and visual diagrams).

• Visual aids and structured visual frameworks are scientifically proven to boost retention and recall, including for people with cognitive impairments. Visual frameworks also provide a concrete and predictable structure that can be particularly beneficial for those who struggle with processing verbal exchanges.

• The markets for speech-to-text apps, speech recognition, and speech analytics are expected to grow significantly in the coming years due to technological advancements.
  

Product Research

There were a few questions I wanted to explore further since their answers could inform the product’s design. This work also served to narrow my focus for user research, so I could be strategic and efficient when interviewing people. Here were some of my insights:

• Is SideTrack a visual working memory? What is working memory, and how does it function, both in the human brain and in computers? Does a shared visual working memory have a place in team collaboration?

  • Working memory allows us to use information without losing track of what we’re doing.

  • It holds new information in place so the brain can work with it briefly and connect it with other information.

  • Its short-term job is to help people tackle the task at hand, but it also helps the brain decide what information to organize for long-term storage.

• Why is context switching challenging? What do I need to know about context-switching to make it more efficient? When working memory is at capacity, what impact does it have on our ability to context-switch?

  • When working memory is near or at capacity, it lowers the threshold for what qualifies as a ‘context’. A context’ then becomes any minute change or shift of attention whatsoever (which is why opening your phone can sometimes feel like walking into a room and forgetting why you went in there).

  • Building up working memory for a task requires a lot of effort. When we switch contexts, we wipe our working memory clean. The more time that passes after clearing, and the more complex the structure was, the harder that working memory session will be to rebuild when resuming the task.

  • People who are experience cognitive overload, or inherently operate with working memory deficits (suggested to apply to those with ADHD, autism, and dyslexia) therefore struggle to see the big picture. Small things are often given the same attention and priority as big things.

• What commonly recognized conversation parameters, facilitation techniques, frameworks, or structures might SideTrack need to visualize? What does meaningful conversation look like?

  • These parameters surfaced most repetitively: goal/purpose, topic, status, importance, relevance, themes, time, clarity, and quality.

  • Meaningful conversation often involves: critical thinking (identifying assumptions and counter-arguments), reasoning and logic (identifying components of the argument, examining the relationship between them, and drawing conclusions.), or persistence (pursuing a line of thought until reaching resolution) — all of which may necessitate returning to previous points when new insights or perspectives emerge.
    

User Research

I strategically selected interview participants who represented our three primary use cases— business, individual, and educational. The objective was to understand their decision-making processes, attention patterns, tool usage, and pain points. Interview questions included:

• What’s the main goal or desired outcome of a [user’s industry] session/meeting?

• How do you prioritize when faced with multiple conversation threads in a [user’s industry] session/meeting?

• What methods or frameworks do you use in a [user’s industry] session/meeting? Are there others that are also widely known and used in [user’s industry]?

  • What are their limitations? Where are they helpful or not helpful? What do your [students / clients / teammates] like or not like about them?

• What tools do you use to keep track of thoughts and ideas in a [user’s industry] session/meeting? Are there others that are also widely known and used in [user’s industry]?

  • How/where did you discover these tools initially?

  • What are their limitations? Where are they helpful or not helpful? What do your [students / clients / teammates] like or not like about them?

• What are your biggest areas of friction in a [user’s industry] session/meeting? Your [students / clients / teammates]’ biggest areas of friction?

• What is your experience with mindmapping and diagramming?

• What are key elements you think an app like this would need to have, in order to be user friendly, for you and your students/clients?

• What ethical or privacy concerns might you or your [students / clients / teammates] have about an app like this?

• Are you interested in being part of ongoing user research for SideTrack?

This approach, coupled with insights from additional secondary research, provided a broad understanding of pain points and needs across different domains.

Persona Development

Three distinct personas emerged, each with unique challenges:

1. Business Professionals (25-50 years): They value productivity and clarity but struggle with maintaining focus on the big picture and long-term goals in meetings.

2. Individual Users (18-35 years): They face recurring decision fatigue, and struggle to balance competing needs and personal values.

3. Educators and Students: They value learning and collaboration but are challenged by the overwhelming influx of information and distractions.

These personas guided every aspect of SideTrack's development, from the website design on Squarespace, to the scripting of our demo video.

Problem Definition

Users who fell into those three categories all shared a mutual problem: Difficulty maintaining a larger perspective— which meant getting lost in the details, and lacking a common understanding. This pattern surfaced most clearly in collaborative or group environments— not just in formal meetings, but in almost any conversation. I was increasingly affirmed in my classification of SideTrack as a visual working memory, since it maintains perspective and context while users work toward their goals. The director of a design agency even said to me, “This would solve all of my problems!”

AI Logic and Development

Initially, we focused on visualizing discussion topics. However, in reality, conversations touch on many topics that aren’t useful in pursuit of a goal. As a result, the prototype created graphs that were cluttered with distracting and irrelevant details. Simply noting topics wasn’t a high enough level of abstraction, so we needed a more targeted approach. It was at this point that I encountered the concept Dialogue State Tracking (DST), which is a process that tracks a user's intentions at each turn of a dialogue.

I decided to revisit research with a new lens in mind. I used an AI research tool called GPT Researcher to first learn what it is, how it works, how to evaluate performance of DST, and whether anyone had attempted to visualize DST. There seemed to be no direct evidence of dedicated research focusing solely on the visualization of DST.

It also dawned on me that in the tech industry, DST traditionally applies to dialogue between humans and AI— but there’s no reason we can’t apply it to interactions between humans and other humans. If we want to use it to our advantage, we need to make it tangible and visible first. I made the decision to alter the AI logic to focus on dialogue state tracking and conversation pivot points rather than topic shifts, and the first test we did resulted in the most useful diagram yet!

Outcome

We've developed an MVP ready for real-world testing, with a growing interest as evidenced by our growing Free Trial waiting list, and commitment from all interviewees to participate in ongoing user research.