Workshops
Rebecca’s Conversation Design Workshop Series
Through the years, I’ve taught hundreds of people in tech how to be conversation designers. If your group is interested in training, I have pre-developed curriculum, and I can create custom courses, too. Happy to teach virtually, or in-person. Either way, I’m there, live, delivering great content, leading useful activities, and responding moment-to-moment to your group’s questions and input.
Bots mimic human conversation. So shouldn’t we understand what they’re trying to imitate? There are key ideas in linguistics that illuminate our work as conversation designers: turn-taking and the speech chain, accommodation, prosody, repair… These basics give you a starting point for your bot.
Linguistics 101
Whether your system is fully LLM-based or hybrid, LLMs aren’t “set it and forget it.” They require the same design, curation, and optimization as classic conversation design. Conversation designers should be primary owners of, or co-stakeholders in prompt engineering to determine what goes into the prompt and how to structure it. Conversation designers are also necessary to evaluating quality and accuracy. I teach LLM basics up to advanced techniques, and can advise on process and production.
LLM Design
Personality Design
Bots need carefully designed personalities — not just to capture a brand’s vibe, but to set user expectations and build trust. The wrong personality doesn’t just feel wrong; it changes how people talk to the bot, and increases the chances of failure. I can help your team create and document a personality that’s consistent, effective, and unique.
Response Writing
Everything a bot says, under every condition, needs to be carefully crafted to be accurate, clear, and concise. And maintain low cognitive load. And capture the right tone of voice. And give the user a clear sign it’s their turn. And provide navigational guidance, but not too much. I distill my years of writing expertise into practical, concrete strategies for creating prompts that work.
Intent Classification
A bad bot doesn’t recognize what people say. A good bot is flexible enough to let people make their request any way they see fit. To make a good bot, you need to understand the variation in how users make requests — and it’s probably more than you think. I can wrangle training data so that it aligns with what people are asking for (and I can teach you how to test it, too.)
Conversations put together what people say and what the bot says. Flow diagrams capture all of the supported pathways that bot can handle, and all the logic that can govern what a bot says and when. In conversation, context is everything, and I can show you how to start to map that context.
Conversational Pathways
Spoken conversations and typed conversations require different design decisions. And the landscape changes even more when you’re working on a multimodal conversation that allows voice, text, and touch inputs. Plus, each type of interface has its own accessibility concerns. I’ll walk you through it.
VUI vs. GUI
When two people talk, conversation repair happens every 84 seconds. It’s a normal and essential behavior. But bots usually drop the ball. I teach a full toolkit of repair strategies, and show you how to understand the kind of repair that’s needed, and how the bot should execute it.
Conversation Repair
Conversation designers have special techniques to test early-stage products, and different methods for mature products. I teach a full range of low- to high-fidelity prototype methods, and best practices for conducting testing sessions.
Prototyping and Process
Every team has different stuff they want to learn. I’m an experienced teacher, so I can put together custom sets of concepts, talks, and activities to target exactly what your team needs.