The business problem behind the build.
The research team needed a way to process earnings calls from more than 500 portfolio companies, surface what changed, and help portfolio managers move quickly when risk or opportunity showed up.
That meant building something more useful than raw transcript search. The interface needed to keep sentiment, playback, and evidence connected in one place.
A system designed for real decision-making.
We built a FastAPI backend to ingest transcripts and audio, plus a React platform that synchronizes sentiment timelines, playback, and transcript highlights.
Audio is segmented for vocal-tone inference, results are cached for fast reuse, and the final experience lets analysts jump straight to the moments that matter instead of scanning whole calls manually.
The decisions that made the workflow hold up.
- Audio sentiment runs on short overlapping windows so teams can see tonal changes over time instead of getting a single blunt score.
- Signals are returned as time-series output and rendered directly in the visualization layer for fast interpretation.
- Playback, charting, and transcript context stay aligned so users can validate why an alert exists.
Workflow View
Earnings Intelligence Pipeline
How call audio becomes research-ready alerts with evidence attached.
Call audio + transcript
The pipeline combines spoken and textual context so models can read the language and detect tonal shifts in executive communication.
Key signal: 500+ companies
Result
The platform became key alert tooling for portfolio managers and researchers, helping them act on earnings-call changes more quickly and with better context.
Because the evidence stays attached to the signal, the workflow supports speed without forcing users to take the model on faith.
What this work says about how ControlThrive builds.
High-value research tooling does not just summarize content. It makes evidence easier to reach at the moment a decision needs to be made.