ChromaScribe
Qualitative data analysis AI tool
This tool was designed to build a relationship between a researcher and an AI. My research revealed that professionals don't want an AI "supervisor" checking their work; they want a transparent "assistant" that handles the grunt work while they handle the thinking. Every screen below was built to enforce that specific hierarchy.
This is ChromaScribe, an AI tool I designed to solve the massive labor bottleneck in qualitative analysis. As the first author of our published study, I investigated the psychological threshold of trust between researchers and machines to define exactly where AI should and shouldn’t intervene.
This setup phase to allow researchers to define the boundaries of their project before a single line is coded. By enabling the upload of specific 'Question Sets,' I ensured the AI has the necessary contextual guardrails to understand the human intent behind the data, preventing the 'hallucinations' that often happen in AI models.
My research uncovered a critical pain point: professionals value non-verbal cues but often ignore them because switching between players is too time-consuming. I designed this pairing interface to link audio and text at the source, effectively lowering the friction of multimodal analysis and capturing more emotional nuance.
A major research finding was that professionals reject AI as a 'supervisor' but embrace it as an 'assistant'. I pivoted our design to an 'AI-Initiated' workflow, where the system suggests candidate themes for the human to approve, preserving the researcher’s interpretive authority and professional identity.
Trust is earned through transparency. I designed a 'Glass-Box' interface that visualizes the specific keywords the AI used to build its themes.
This allows for instant auditing, ensuring the researcher is always the final decision-maker and remains 'in the loop' of the analysis.
Note: Transcripts have been blurred due to data sensitivity.
This is where the impact happens. This interactive timeline provides a high-density map of themes across 50+ hours of data. By syncing every thematic block directly to the audio and transcript, we validated an increase in analytical efficiency, turning a four-hour task into a two-hour workflow.
Note: Transcripts have been blurred due to data sensitivity.
To further support the researcher’s mental model, I added a strategic pivot: Question View. This allows the user to re-orient the entire dataset by specific interview goals, surfacing hidden patterns that standard linear tools consistently overlook. ​​ChromaScribe proves that when we design for human ownership, AI becomes a powerful catalyst rather than a replacement.
