Can a prompt replace Read AI?

Build a meeting-analytics tool computing basic speaking-time and sentiment stats from a transcript, on top of this catalogue's Avoma-style local transcription approach. Use Python with FastAPI, whisper.cpp for transcription with speaker diarization, manual correction, same approach as the Avoma entry, and a lightweight open sentiment-analysis model, a small HuggingFace transformer model run locally, for per-segment tone scoring. After transcribing and speaker-labeling a recording, compute: total speaking time per participant as a percentage of the meeting, a simple timeline chart of sentiment score over the meeting's duration, and a basic 'most positive' and 'most negative' segment callout. Present these as a summary dashboard alongside the transcript. Be explicit in the README that this is a lightweight, unvalidated heuristic, not a scientifically validated engagement model — sentiment analysis on transcribed speech, missing tone-of-voice nuance, is genuinely imprecise, and presenting it as more authoritative than it is would be dishonest. Do not build calendar-triggered automatic recording, cross-meeting analytics over time, or video-based engagement signals — those are out of scope. Needs hosting to run continuously, or can run on a personal machine for occasional use; no external API key required for the local sentiment model.