Legal teams rely on transcripts for decisions that carry real risk, but most transcription tools optimize for speed over trust. At CaseMark, transcription wasn't just a utility feature. It was the foundation for summaries, analysis, and downstream legal work. The challenge was to turn raw audio into text that users could confidently verify, correct, and rely on—without slowing them down.
Problem
- →Raw AI transcripts were difficult to verify against audio
- →Speaker attribution broke down in crowded or messy recordings
- →Users had no clear signal for where transcripts were likely wrong
- →Reviewing long transcripts required too much manual effort
Goals
- →Make transcripts easy to verify against audio
- →Design for correction, not perfection
- →Help users focus attention where accuracy matters most
- →Support legal workflows without adding complexity
Solution
Playback synced to text
Audio playback is tightly synced to the transcript so users hear and see the same moment at once. Words highlight as they're spoken to support quick verification. When speaker pauses caused visual jumps, I eased the highlight transition so playback felt stable and predictable. Keyboard controls and clickable timestamps support fast, repetitive review.
Speaker identification built for uncertainty
Speakers are identified automatically when confidence is high and left unassigned when it's not. Users can correct speakers inline or manage them from a central list. This avoids false attribution and keeps corrections fast in crowded or imperfect recordings.
Confidence signals and inline editing
Each word surfaces a confidence signal so users know where to focus review. Low-confidence words draw attention while high-confidence text fades into the background. Inline editing keeps users in context and reduces unnecessary rereading.
Versioning aligned to legal work
Versions are created at meaningful milestones: original transcript, edited transcript, and derived outputs. This provides a clear audit trail without tracking every keystroke and mirrors how legal work actually progresses.
Impact
The workflow shifted transcription from a passive output to an active review tool. Users could verify accuracy faster, correct issues with confidence, and trust transcripts as a foundation for downstream legal work. The design reduced friction without hiding uncertainty and supported real-world legal review patterns.