
We've all been in that meeting where everyone agrees on what needs to happen, and then nothing gets logged. Someone has to go back through their notes, figure out what was decided, write it up as a ticket, guess at the priority, and hope they got the assignee right. It's tedious and it breaks down constantly. We built Meeting2Jira to handle that step automatically. You upload a meeting transcript — a .vtt file from Teams or Zoom, or a .docx — and the app sends it through IBM watsonx.ai using llama-3-3-70b-instruct. The model pulls out action items with title, description, assignee, priority, due date, and the exact quote it based the ticket on. Not every extraction is equally reliable. If the model isn't confident — maybe the assignee was implied, or the task was vague — we flag it with an amber warning rather than quietly pushing a bad ticket to Jira. Anything under 70% confidence gets flagged for human review. We'd rather surface uncertainty than pretend everything is fine. The UI mirrors Jira's design so switching between the two doesn't feel jarring. Review the tickets, edit anything that's off, hit push. They show up on your board with the right issue type, priority and assignee. We built the whole thing using IBM Bob IDE. Bob handled the backend scaffolding, watsonx.ai integration, Angular components, and the Jira API client — across TypeScript, Express, Angular, and two external APIs simultaneously. It also works in German and French. We tested it with transcripts in both and the extraction held up well.
17 May 2026