The bot that keeps the project alive.
Delphina started as a Jira worklog assistant — something to close the gap between shipping code and recording the hours. It grew: daily summaries, weekly reports, a read-only PM portal, and a pipeline of capabilities being wired in (email triage, CloudWatch watching, bug intake, GitHub commit matching).
The goal is ambient awareness. Delphina should tell you things before you think to ask.
Running in production.
- Worklog submit — type a structured message in Google Chat → Delphina posts to Jira with human vs AI hours split
- CSV drop — bulk worklog import from timecard exports
- Daily report — open tickets + worklogs posted to Chat each morning
- Weekly summary — full worklog digest with hours breakdown
- PM portal — read-only API for project managers to pull current state
- Multi-tenant — Jira credential resolution per developer, one bot serves the team
Phase 3 capabilities.
- Email triage — 7-phase pipeline: fetch → classify → compare docs → draft reply → queue for PM review
- Log watcher — CloudWatch → error detect → auto-create Jira ticket → Google Chat alert
- Bug intake — natural language bug report → log comparison → Jira issue creation
- GitHub worklog suggest — daily commit scan → Jira match → suggestion cards for dev approval
Lambda on Quasslabs AWS. AgentCore next.
Delphina runs as 6 Lambda functions on the Quasslabs AWS account (us-east-1), triggered by API Gateway webhooks and EventBridge schedules. State lives in 17+ DynamoDB tables. Migration to AWS AgentCore is in progress — the architecture decision was made in May 2026 to move the worklog + report agents onto AgentCore for native memory and Bedrock-first tooling.