From meeting speech to task dispatch: wiring on-device ASR into an AI collaboration workflow

Description and use

An end‑to‑end pipeline that captures meeting speech on device, extracts action items, creates Octo Loops, assigns them to agents, and iteratively refines deliverables through feedback.

Problem

Roughly 40% of verbally agreed tasks in daily meetings never made it into any tracking system, leading to missed action items.

Outcome

Implemented an end‑to‑end pipeline that captures spoken action items, automatically creates task Loops in Octo, assigns them to agents, and iteratively refines deliverables through feedback, significantly improving task capture and reducing manual follow‑up.

Workflow

• Record audio with Octic hardware
• Run on‑device ASR to produce timestamped, speaker‑attributed JSON transcript
• Post‑meeting batch NLU extracts action items, decisions, and open questions
• Map extracted action items to Octo Loops via API, setting title, owner, acceptance criteria, and source context
• Notify assigned Agent in Octo, who executes the task and submits deliverable
• Human reviews deliverable, accepts or rejects, providing feedback stored as Preferences
• Agent revises based on feedback, iterating until acceptance

Evidence

Roughly 40% of verbally agreed tasks never made it into any tracking system.

Source

The output is a timestamped, speaker‑attributed transcript in JSON.

Source

We run a lightweight NLU pass over the full transcript.

Source