Inspiration
What it does
How we built it
Challenges we ran into
Accomplishments that we're proud of
What we learned
Inspiration
Engineering teams worldwide spend over 35% of their working hours reproducing, triaging, and writing mechanical fixes for recurring runtime bugs (like ZeroDivisionError, KeyError, IndexError, and null-pointer exceptions). Even when Jira tickets contain pristine stack traces and reproduction steps, engineers are still forced to manually context-switch, locate faulty lines, write patches, and run regression tests.
We built JiraAutoHeal AI to transform Atlassian Jira from a passive issue tracker into an autonomous, self-healing engineering ecosystem powered by Atlassian Rovo and Forge.
What It Does
- Automated Ticket Ingestion: Listens to Jira Webhooks for new or updated bug tickets, extracting stack traces, environment variables, and reproduction steps.
- Semantic AST Codebase Mapping: Parses Python/JavaScript AST trees to correlate runtime error line numbers with exact functional definitions and call sites.
- LLM Patch Synthesis: Synthesizes minimal, robust patches that resolve root causes without breaking existing business logic.
- Isolated Test Sandbox Verification: Executes regression tests before and after the patch inside an isolated sandbox subprocess. If tests fail, the agent iterates; if they pass, the patch is locked.
- Jira Status Auto-Transition & Audit Trail: Updates ticket state to Resolved, appends human-readable git diffs and test logs as audit comments, and reassigns to the lead for final sign-off.
How We Built It
- Agent Core & AST Engine: Python 3.10+ utilizing
ast,subprocess,tempfile, and isolated sandbox test runners. - Interactive Web Interface: Modern dark-mode responsive dashboard styled with Tailwind CSS, live issue backlog, code diff viewer, and instant patch trigger buttons.
- Production Edge Hosting: Live globally at https://jira.elnemer-ai.com powered by Hostinger cloud edge with HTTP/3 and SSL.
- Atlassian Integration: Designed to leverage Atlassian Forge UI Kit and Rovo Agent action triggers.
Challenges We Overcame
- Preventing Hallucinated Patches: LLMs can generate plausible-looking patches that break existing tests. We enforced a strict Pre/Post Execution Sandbox Gate: a patch is automatically rejected unless 100% of existing regression tests pass.
- Sub-Second Execution: Optimized AST line-index mapping to pinpoint failure locations within milliseconds rather than traversing entire repositories.
Accomplishments That We're Proud Of
- Verified 0.6-second resolution time on production bugs (like zero division in cart computation).
- Full auditability with clear, human-readable Git diffs posted directly to Jira.
- Live production dashboard accessible worldwide at https://jira.elnemer-ai.com.
What's Next for JiraAutoHeal AI
- Direct integration with Bitbucket and GitHub PR creation APIs.
- Multi-file patch synthesis for complex architectural refactoring. ## What's next for JiraAutoHeal AI: Self-Healing Bug Agent