We began brainstorming concepts based on our research findings, and found ourselves particularly drawn to how we could leverage existing peer sharing through popular video sharing models, like Vine or Snapchat, or familiar formats, like hackathons. Inspired by these seed ideas, we came up with six broad directions to explore:

• Hackathon
• Role playing
• Practice simulation
• Automatic feedback
• Peer feedback
• Campfire stories

We developed the most compelling of these concepts into eight high-level storyboards. Here is a selection of the ideas generated:

We presented our storyboards to students, faculty, language experts, and educational technologists during speed dating sessions, collecting their feedback and assigning a rough score to each concept. Patterns emerged in the general sentiments towards each design idea; the most popular ideas were ones that incorporated the following:

• Virtual reality: An exciting new medium
• Automated feedback: Quick turn-around, data visualization to track progress
• Multiple perspectives: A mix of crowd-sourced opinions that cover all bases

In addition to gathering feedback about preferred formats, the speed dating process also illuminated several values when it came to practicing presentations and receiving feedback:

Low-level feedback is valuable

Although the benefits of high-level tips for plotting rhetorical strategy can't be overstated, alerting students to more technical painpoints like volume and eye contact still has a positive effect on improving their communication skills.

Students need a solution that works around their schedules

Ideas that required significant time commitments and worked on structured schedules were less popular than those that could be accessed at anytime.

Peer feedback was welcome as a complement to expert feedback

Students were interested in their peers' opinions, since they often shared the same curricular interests and could speak to the clarity of the presentation. Educators also agreed that engagement in a communication skills critique provoked an increased awareness in a student's own speaking style.

As a result, we decided to move forward on a concept that would both allow students to practice on their own time and automatically generate feedback. To retain valuable human input, this concept would work in conjunction with an in-class format that supported crowdsourced peer feedback through a low-friction survey, as well as opportunities for expert feedback from educators.

In order to provide a high-fidelity version of this service concept, we developed a simulator for live practice with a virtual audience. Users were invited to a designated practice room and asked to give a short, informal presentation on a topic of their choice while team members operated the prototype with the "Wizard of Oz" technique, providing live feedback under the guise that it was being generated automatically. The system measured quantifiable rhetorical factors such as pace, volume, body movement, eye contact, and the use of filler words. The live nature of this feedback would prompt students to modify their behavior in real time. We also made a template for a feedback form that we could fill out immediately and show to participants.

We prototyped the practice room experience and the in-class feedback scenario with six students, briefing them on the simulation first and asking them to think of a topic. Then they presented while we recorded metrics and critique. We quickly entered this data into a mocked-up feedback report for the participants, then discussed the experience.

Feedback from the user enactments both equipped us with information about how to improve our prototype and highlighted its most successful components. Students found the live indicators helpful with the exception of filler words, which were often difficult to suppress: They felt flustered when they saw words like "um" or "like" surface repeatedly. Instead we resolved to display only the feedback that could be controlled with relative ease, like volume and speed, and left filler word statistics for the automated report.