Abstract

A method is disclosed for starting several AI voice participants into a scheduled multi-party meeting once per meeting occurrence, even though a periodic scheduler runs repeatedly, the calendar holds duplicate records of one meeting and launches can fail ambiguously. Duplicate records are grouped by attendee list, occurrence instant and title. For each attendee with a known identity and an execution environment, the scheduler first acquires that participant's exclusive-engagement lock, and only then writes a run-once record keyed by meeting record and occurrence, never by participant. If no attendee could be locked, nothing is written, so a later invocation inside a grace window retries with the run-once key unspent. A record in the failed state is re-claimed; if the record exists in any other state, the locks just taken are released. The locked attendees are launched concurrently, each releasing its lock when its launch ends, and the record's status aggregates the launched attendees' outcomes into joined, launched-detached, partial or failed, with busy and unlaunchable attendees itemized in its detail. Each launch attempt carries a fresh attempt identifier that the remote launcher writes into the execution environment in the same step that wipes the previous session's log. After a failed attempt the scheduler probes the environment and trusts a join marker only when the stamped identifier equals the current attempt's, so a stale marker cannot report success. When a join-verification window is configured, an alive client that has not joined gets that long to join or is terminated before any relaunch.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Share

COinS