An AI-powered platform that transforms Jira sprint data into HTML newsletters for stakeholder communication, with structured prompts and per-request logging.
Status
Active
Role
Full-stack engineer responsible for the Go service, Next.js workflow, and LLM orchestration
Duration
2024
Ownership
End-to-end engineering
Problem and constraints
Engineering teams needed to turn detailed Jira sprint activity into concise stakeholder communication without repeatedly copying issue data, manually restructuring it, and editing inconsistent model output.
Private enterprise project
Jira authentication
LLM token limits
Email-compatible HTML
System flow
01Select a board and sprint through Jira
02Curate the issues included in the report
03Normalize Jira content and assemble a constrained prompt
04Generate and clean email-ready HTML
Engineering decisions
Curate before generation
Manual issue selection keeps irrelevant Jira activity out of the prompt and gives the operator control over the report scope.
Constrain the output contract
Canonical sections, HTML-only instructions, and cleanup make the generated result more predictable for email clients.
Evaluation
Report quality
Confidential. Reports were reviewed for issue fidelity, required section coverage, usable HTML, and absence of unsupported additions.
Reliability, safety, and observability
Pre-flight token validation warns before oversized requests.
Per-request logs support debugging and review; client timeouts prevent indefinite generation states.
Outcomes
Automated the Jira-to-newsletter generation workflow
Produced email-ready HTML through a guided interface
Failures and lessons
Human curation before generation improves relevance and reduces prompt size.
Structured output still requires normalization and validation at the application boundary.
Evidence
Implementation details and source code are not public because this is a private repository.