1. Overview
In enterprise banking ecosystems, diagnosing intermittent system exceptions often required senior engineers to manually cross-reference proprietary runbook PDFs, error codes, and source code across multiple private repositories.
The 5-Minute RCA Tool brings playbook matching and repository search into one investigation flow. It proposes a cause and relevant references for an engineer to verify.
2. The Problem
During high-severity incidents, manual triage suffered from high cognitive friction:
- Error messages like generic null-pointer or timeout exceptions failed to point to the upstream offending service.
- Runbooks were scattered across Confluence and internal PDF repositories.
- On-call engineers spent valuable minutes grepping repository code trees across microservices.
3. Solution & Workflow
4. Technical Decisions
- Zero-Dependency Lightweight Backend: Engineered in standard Python without heavy orchestration frameworks, ensuring rapid startup and zero operational baggage.
- Evidence-first output: The diagnostic prompt asks for file and playbook references and should return no match when evidence is insufficient. An engineer still checks the proposed cause.
- Security Guardrails: Sensitive credentials, API keys, and customer data are scrubbed before analysis.
5. Impact & Results
- Combines recurring manual search steps in an internal engineering workflow.
- Returns a structured briefing with candidate cause, possible fix, and references for review.
Evaluation & limits
The public case study omits proprietary examples and has no published time-savings benchmark. Code search can miss a cause outside the indexed repositories; proposed fixes require engineer verification.
A useful next step is to compare diagnoses with resolved incidents, tracking helpful matches and false leads.