To export your project:
- Go to Settings > Integrations > Export Options
- Toggle PDF Export to On
- Open any project and click the Export button in the toolbar
- Select PDF from the dropdown
Note: PDF export is available on Pro and Enterprise plans only.
Context
What Does This Data Actually Look Like?
5 minBefore we work with customer support conversations, let's see what we're actually talking about. These are the kind of conversations and knowledge base articles you'll be working with throughout this course.
All our breakout exercises use customer support conversations as a running example. Before we start, let's look at what these conversations actually look like — because "conversations" is deceptively simple.
Example conversations
Conversation A
Conversation B
Conversation C
Conversation D
Example knowledge base articles
Knowledge Base
How to Export Projects as PDF
Knowledge Base
Troubleshooting Slow Dashboard Performance
If your dashboard is loading slowly, try these steps:
- Clear browser cache and cookies
- Disable browser extensions (ad blockers can interfere)
- If you have 100+ projects, use Workspaces to organize them — this improves load time
- Try a different browser to rule out browser-specific issues
If the problem persists, contact support with your browser version and a screenshot of the Network tab in developer tools.
Notice: Conversation A maps cleanly to a KB article. Conversation B spans billing, access, and churn risk. Conversation C starts as performance but is really about organization. Conversation D isn't a bug at all. Now imagine clustering 8,000 of these "by topic." What does "topic" even mean here?
- Walk through these quickly — the point is to build intuition for the data complexity, not to analyze each one deeply.
- Key moment: Conversation C (topic drift) and D (not a real support issue). These make students question what "clustering by topic" means.
- Optionally ask: "If you had to label each conversation with ONE topic, what would you pick for Conversation B?" — students will disagree, which is exactly the point.
Setup
The Scenario
2 min
A company has shared a large corpus of customer support conversations with your research group. Four different people in that company each have a different problem they need help with. Your group will meet all four of them. For each one, formulate a research question that addresses their specific need — and give it a name. Same data, same tools — but different needs lead to different research.
Shared Context
Data
50,000 customer support conversations (text)
Metadata
Timestamps, agent ID, resolution status
Avg conversation
12 messages, 2 participants
Domain
SaaS product (project management tool)
Time span
18 months
Tools
Embedding models, clustering algorithms, 500 human labels max
Constraint
3-month timeline · Deliverable: research report + prototype
Context
Four Perspectives, Four Pain Points
Each of these people works at the same company and has access to the same conversation data. They each need something different from it.
Stakeholder A
Dana Chen
VP of Support Operations
"My agents spend 40% of their time just figuring out what the problem is before they can start helping. We have 50 agents handling 200 tickets/day. If I could auto-route tickets to the right specialist, I'd save $2M/year. I don't care how you do it — I need tickets going to the right person within 30 seconds."
Success: Faster resolution, fewer transfers between agents.
What research question addresses Dana's need? Give it a name.
Stakeholder B
Marco Silva
Head of Product
"I have no idea what our users actually struggle with. Product decisions are based on gut feel and whoever shouts loudest in the Slack channel. I need a systematic way to understand: what are the top pain points? Are they changing over time? When we ship a fix, does the conversation pattern actually change?"
Success: "I can point to data when I prioritize the roadmap."
What research question addresses Marco's need? Give it a name.
Stakeholder C
Priya Kapoor
Chief Risk Officer
"Last quarter we had a billing bug that affected 2,000 users. Tickets started trickling in on Monday, but we didn't notice until Thursday when it hit social media. By then we had 500 angry public posts. I need an early warning system. I need to know when a NEW type of problem is emerging before it becomes a crisis."
Success: Detect emerging issues at least 48 hours before they escalate.
What research question addresses Priya's need? Give it a name.
Stakeholder D
Alex Novak
ML Research Lead
"I'm interested in whether current embedding models actually capture the pragmatic structure of multi-turn conversations, or whether they just encode surface lexical features. Conversations have structure that single sentences don't — turn-taking, topic drift, resolution patterns. I want to understand what clustering reveals about representation quality. Publication is the goal."
Success: Novel finding about conversation representations, publishable at a top NLP venue.
What research question addresses Alex's need? Give it a name.
Phase 1
From Need to Question
30 minThese four "personas" show four different pain points from the same data. Your job: define a research project — give it a name and formulate research questions. Your project might address one stakeholder's need or cut across several. Not "clustering" (that's a technique, not a project). If you can't name it crisply, your questions probably aren't specific enough yet.
Group output — one sheet of paper per group
- Project name — 2–6 words that capture what your research is about
- Research questions — specific, answerable questions. A single project may address one stakeholder's need or several.
Project these questions
- "What is the actual problem this person faces?"
- "What would change in practice if your research succeeds?"
- "How could you be wrong about what this person actually needs?"
- "Which are the interesting and useful problems that — if solved, or even if asked — would advance knowledge and lead to useful outcomes?"