Cut Time: AI-enabled qualitative research for assistive tech

The Dari project shows that aligning transcripts, caption streams, and EMG signal logs produces actionable priorities for assistive communication products. Evidano ingests those multimodal inputs and accelerates thematic, frequency, and cross-segment analysis so teams can move from data to decisions in hours rather than weeks. This post summarizes the Dari findings, the data types you will see, and a 7-step checklist to pilot a Dari-style study.

• Dari combined three components (Woori EMG armband; Sari smart glasses; companion app) during a 26-week design process reported May 8, 2025.
• Key research outputs to align are interview transcripts, time-stamped caption text, and EMG-derived event logs to preserve conversational nuance and measure latency.
• Evidano ingests multimodal corpora, aligns timestamps, runs automated thematic coding, and produces cross-segment reports and exportable codebooks.

The Dari project shows that a three-part assistive communication ecosystem surfaces adoption barriers, privacy trade-offs, and UX priorities that require multimodal analysis. Dari is an assistive communication ecosystem that aims to bridge Deaf-hearing conversation with three integrated components: Woori (EMG armband), Sari (smart glasses captioning), and a mobile companion app. The project was developed over a 26-week design process and reported publicly on May 8, 2025, in Core77.

• Problem: fragmented inputs (interviews, EMG signals, caption text) and a need to preserve conversational nuance.
• Opportunity: combine qualitative interviews and multimodal logs to surface adoption barriers, privacy concerns, and UX trade-offs.
• Payoff: actionable priorities (device accuracy, gaze-preserving caption UX, offline interpreter fallbacks).

This findings snapshot summarizes the project timeline, core components, primary methods, and design goals identified in the public report.

This section explains how Dari works in plain English by describing its three core components and their data outputs. Woori is an EMG armband that captures forearm muscle signals and uses machine learning to infer sign gestures without cameras, reducing environmental constraints and privacy concerns. Sari is smart glasses that display real-time captions in the user's field of view so Deaf users can keep eye contact and see conversational context. The companion app provides device management, personalization, communication history, and settings that stitch wearable outputs with conversation logs.

• Data outputs you will see in a research corpus: interview transcripts, annotated caption text (time-stamped), EMG-derived event logs, usability notes from co-design sessions.
• Analytically relevant problems: aligning timestamps across modalities, coding non-verbal cues, measuring perceived conversational flow versus objective latency.

This checklist gives a seven-step pilot to reproduce Dari-style insights using aligned multimodal data.

Step 1: Collect raw inputs, interviews, SRT caption files from Sari prototypes, EMG logs from Woori (timestamped).

Step 2: Import to Evidano (Evidano) and align timestamps across modalities.

Step 3: Run automated transcription normalization and apply a custom dictionary for sign labels.

Step 4: Apply initial codebook (importable) and run AI-assisted coding to generate preliminary themes.

Step 5: Produce cross-segment reports (location, device confidence, participant persona) and flag high-impact quotes.

Step 6: Validate with 2–3 participant reviewers and update codes; keep versioned exports for audit.

Step 7: Deliver a one-page decision memo with prioritized product changes and metrics (latency reduction, error hotspots, adoption risk).

Dari demonstrates how multimodal assistive systems generate research challenges that are solvable with integrated qualitative analysis by aligning transcripts, captions, and sensor logs to preserve conversational nuance while producing evidence product teams can act on. Ready to reproduce this workflow? Try Evidano for free and upload your interview transcripts, caption files, and EMG logs to run thematic, frequency, and cross-segment analyses in hours.