Argo | Devpost

Logo
Changes Made
Live Meeting List
Transcript
Commands Executed
Meeting Details and Ticket Creation
Meetings List
Meeting Summary and Key Points
Meeting details and transcript
Commands executed
Jira Board Updated by Argo
Argo talking
Argo Listening
Argo waiting
Argo Stopped

Inspiration

We've all sat through scrum meetings where good decisions never turned into follow-ups, because the tickets didn't get created, updated or assigned afterward. We wanted to save scrum masters that time and make follow-ups impossible to forget by letting the team handle them right in the middle of the meeting.

What it does

Argo is a voice agent that listens in on your standup or sprint meeting and acts on Jira commands that start with "Hey Argo." You can ask it to create, update, assign, transition, comment on, search for or delete issues, and to move work between sprints and the backlog. It can chain several of these actions from one request, and it replies out loud so the team knows the change went through. Every meeting's transcript, commands and Jira changes are saved, and when the meeting ends Argo summarizes it and shows the whole history inside Jira.

How we built it

The core is a Python agent that streams meeting audio to Cartesia's ink-2 speech-to-text, listens for the wake phrase, and sends each command to Claude. Claude turns the messy spoken command into a structured, validated plan of Jira REST operations, which our executor then runs against the Jira Cloud and Agile APIs. Meetings, transcripts and operation logs are stored in a SpacetimeDB database that serves its own REST API. Inside Jira, an Atlassian Forge app shows that history on a global page and on a panel attached to each issue, and a native macOS app lets users start and stop Argo without a terminal.

Challenges we ran into

Our biggest challenge was getting the agent to listen properly: it had to tell when someone was actually speaking to it and when they had finished, even with the natural pauses of a live meeting. Just as hard was making sure Claude turned loosely worded requests like "move scrum dash twelve to done" into exactly the right Jira API calls.

Accomplishments that we're proud of

We're proud that Argo works in essentially real time, with Jira changes going through within seconds of a command being spoken. It also handles a wide range of Jira actions, including several in a row from a single request, such as finding a ticket by name, reassigning it and moving it into the current sprint.

What we learned

We learned how hard it is to reliably turn unstructured speech into structured actions. Speech-to-text output is noisy, so we built a strict operation schema and validation layer and taught the model to resolve tickets by search and pass results from one step to the next, rather than guessing ticket keys. That taught us that a reliable AI agent depends as much on the guardrails around the model as on the model itself.

What's next for Argo

We want to make Argo a full Atlassian Marketplace extension and bring it directly into Zoom and Microsoft Teams, so it joins meetings without being run separately. We also plan to have Argo infer action items from the conversation and suggest Jira updates on its own, so even unspoken follow-ups get captured.