What I learned — and where this could go next
This project taught me far more than I expected.
What started as an idea for reusing knowledge from resolved incidents evolved into a complete end-to-end system combining CockroachDB vector memory, AWS, Amazon Bedrock, ServiceNow, GitHub Actions, security controls, testing, and an operational demo workflow.
Even producing the final demo video became a challenge of its own — building the story, recording the full workflow, preparing narration, editing everything together, and explaining the technical value in under three minutes.
More than a ticket recommendation engine
What excites me most is not only what the project does today, but what this architecture could unlock in the future.
Today, Agentic Incident Memory can retrieve relevant resolved incidents and provide grounded, explainable recommendations directly inside the incident workflow.
The same foundation could eventually support much more proactive operational capabilities:
- continuously analyze GitHub Actions or application logs
- detect recurring failure patterns automatically
- correlate new failures with verified historical incidents
- surface remediation guidance before an engineer starts searching manually
- evolve toward self-health and automated recovery workflows
The important part is that this does not have to remain just an AI assistant that suggests similar tickets.
It can become a persistent operational memory layer that learns from verified outcomes and helps systems understand what worked before.
The biggest takeaway
Building the architecture was challenging.
Connecting all the components into one coherent workflow was even harder.
And presenting the entire idea clearly in a short demo turned out to be another engineering problem on its own.
But that is also why this has been one of the projects from which I have learned the most.
Remember what worked. Trust what was verified.
Log in or sign up for Devpost to join the conversation.