Glór (Voice)

Glór is a multi-agent meeting intelligence platform that transforms raw audio recordings into structured, actionable meeting outputs. Users upload a meeting audio file directly from their device - the system uploads it to AssemblyAI, transcribes via its async API with diarisation, extracts action items, summarises, analyses sentiment, detects speaker conflicts, and scores meeting quality on a1-10 scale - then distributes personalised follow-ups via Gmail and Slack in parallel, all gated by a human review step. A persistent accountability layer tracks action items across meetings. A Demo Mode injects a curated sample transcript (with deliberate conflicts, action items, and circular discussion) directly into the analysis pipeline, skipping transcription entirely.

The multi-speaker transcription and diarisation layer in Glór
builds on academic research conducted as part of my MSc in
Artificial Intelligence at National College of Ireland (thesis
in active development, March–August 2026). My thesis focuses on
context-aware real-time speech-to-speech translation with
multi-speaker support, using pyannote.audio for speaker
diarisation and OpenAI Whisper for transcription — the same
core technology stack underlying Glór’s Transcription &
Diarisation Agent.

No prior code was reused directly. The agent architecture,
multi-agent orchestration pipeline, accountability memory layer,
conflict detection logic, meeting quality scoring system, Gmail
and Slack integrations, database schema, and full application
were designed and built entirely during this hackathon using
Lyzr Architect.

The academic background in speaker diarisation informed the
design decisions around the AssemblyAI integration —
specifically the three-step async flow, speaker label mapping,
and utterance formatting — and allowed this component to be
architected correctly and quickly where other teams might have
struggled.