Brushstrokes & Lies

Critical thinking doesn’t happen in a vacuum — learners need a reason to start asking questions before they’re handed any information. So I began the experience with a short animated sequence showing a shadowy figure swapping the painting and fleeing the scene, rather than opening with a briefing or a list of facts. Starting with the crime itself creates immediate curiosity and prompts learners to form their own questions before Detective Ruiz tells them anything.

After the title screen, Detective Ruiz — a senior detective overseeing the investigation — explains what happened, introduces the two suspects, and notes that additional evidence is still being collected. She also introduces a notebook learners can use to record observations, a small structural cue that this is an active investigation, not a story being told to them. From there, learners reach a suspect-selection screen and choose whom to question first. That choice was intentional: real investigative reasoning starts with deciding where to look, not just how to respond once you’re told where to look.

The central interaction consists of two AI-powered suspect interrogations built with devlin.ai and embedded in Articulate Storyline 360. Learners can ask each suspect up to ten original questions rather than selecting from multiple-choice prompts. This open-ended structure is central to the learning: learners must determine which topics are relevant, phrase useful follow-up questions, recognize evasive or suspicious responses, and decide when they have enough information to move forward.

One technical challenge was making the two interrogations feel like parts of the same investigation, since the AI simulations operate independently and can’t see what happened in the other conversation. To create continuity, I gave each simulation shared scenario context — knowledge of the gallery, the original painter Elena, the other suspect, and the suspicious behavior surrounding the incident — so each conversation references the same world and events even though the two AI chats never communicate directly.

After both interrogations, Detective Ruiz returns with up to four pieces of evidence. Elena’s police statement is always available, but the remaining evidence depends on what the learner investigated during the conversations. One piece can be uncovered through questioning either suspect, while one is tied specifically to Julian’s interrogation and another to Adrian’s.

This conditional reveal reinforces the importance of asking effective questions. Evidence isn’t simply handed to the learner after the chats — their questioning strategy determines whether key information becomes available. It also creates a meaningful consequence for missed opportunities: a learner may suspect the correct person but still lack enough evidence to build a defensible case.

Once learners review the evidence, they must select the suspect they believe is responsible. The next scene takes place in the district attorney’s office. When the learner has uncovered enough evidence, the district attorney thanks them and agrees to move forward with the case; when the investigation is incomplete, he explains that there isn’t enough to proceed.

This distinction was intentional. The learner’s task is not merely to guess the correct suspect — they must conduct an investigation strong enough to support their conclusion. The outcome then plays out through a newspaper-style scene before the learner receives personalized feedback.

Because every learner can ask different questions, generic feedback wouldn’t adequately reflect their performance. I used devlin.ai to generate feedback based on the learner’s actual conversations, focusing on how effectively they asked relevant questions, explored suspicious details, followed up on inconsistencies, gathered supporting evidence, and used that information to reach a conclusion. Rather than simply telling learners whether they were right or wrong, the feedback helps them reflect on the quality of their investigative process.