Inspiration
What it does
How we built it
Challenges we ran into
Accomplishments that we're proud of
What we learned
What's next for Ops-Copilot: AI-Powered Incident IntelliHere's a polished Devpost "About the Project" section based on your project and README.
🛡️ OpsCopilot — AI-Powered Incident Investigation Platform
Inspiration
Modern DevOps and Site Reliability Engineering (SRE) teams rely on multiple monitoring tools such as CloudWatch, Grafana, deployment dashboards, and documentation portals to diagnose production issues. During incidents, engineers spend valuable time switching between these tools, manually correlating logs, metrics, deployment history, and documentation before they can even begin identifying the root cause.
We wanted to simplify this workflow by building an AI-powered operations assistant that performs the investigation itself instead of simply answering questions. The goal was to create a platform that reduces investigation time, improves operational efficiency, and helps engineers move from incident detection to resolution through a guided, evidence-based workflow.
What it does
OpsCopilot is an AI-powered incident investigation platform that transforms natural language questions such as:
"Why are payments failing in production?"
into a complete incident investigation.
The platform first analyzes the request using an AI Investigation Planner that determines the user's intent, creates an investigation strategy, and coordinates specialized analysis modules to gather logs, deployment information, metrics, and operational context.
The collected evidence is fused into an executive summary containing probable root causes, severity assessment, confidence score, and recommended actions.
With a single click, engineers can continue into the Incident Copilot, where a multi-agent workflow guides them through:
- Evidence Correlation
- Log Analysis
- Root Cause Analysis (RCA)
- Response Plan Generation
- Incident Command
- Action Execution (human-approved)
- Postmortem Generation
Rather than acting like a chatbot, OpsCopilot behaves like an intelligent incident investigator.
How we built it
The frontend was built using React 18 and Vite, providing an interactive dashboard and modular investigation interface.
The backend uses FastAPI, while Google Gemini 2.5 Flash powers the reasoning capabilities throughout the application, including:
- Intent understanding
- Investigation planning
- Evidence summarization
- Root cause analysis
- AI recommendations
- Postmortem generation
We designed a custom orchestration engine that plans investigations, executes multiple analysis modules, collects their outputs, and fuses the evidence into a single grounded response.
To make demonstrations realistic, we built a live payment service capable of injecting failures such as database pool exhaustion, gateway timeouts, and latency spikes. This produces realistic logs, metrics, and deployment events that OpsCopilot analyzes in real time.
The application also includes documentation drift detection, platform setup guidance, AI-powered log exploration, response planning, and contextual operational recommendations.
Challenges we ran into
One of the biggest challenges was ensuring the AI remained grounded in evidence instead of generating plausible but incorrect explanations.
We addressed this by introducing explicit grounding rules that require every AI-generated conclusion to be supported by collected evidence. When the system detects that services are healthy, it deliberately reports that no incident exists instead of inventing one.
Another challenge was coordinating multiple AI modules while maintaining shared context across the investigation pipeline. We implemented a shared global context that automatically carries investigation results between modules, eliminating repetitive user input and ensuring consistency throughout the workflow.
Designing a realistic incident simulation environment was also challenging. We developed a live payment service capable of injecting production-like failures, allowing the platform to demonstrate end-to-end investigations on dynamic operational data.
What we learned
Throughout this project, we gained practical experience in designing AI agent workflows, prompt engineering, evidence-based reasoning, and multi-stage orchestration.
We also learned the importance of grounding AI responses with operational data to improve trustworthiness, reduce hallucinations, and support engineering decision-making.
Building the investigation pipeline reinforced how AI can augment—not replace—human operators by providing context, recommendations, and structured investigations while keeping engineers in control of operational decisions.
Accomplishments that we're proud of
- Built an end-to-end AI-powered incident investigation platform.
- Designed a multi-stage investigation workflow that mirrors real SRE practices.
- Developed a custom orchestration engine for coordinating multiple AI modules.
- Created a live failure injection environment for realistic demonstrations.
- Achieved 100% RCA accuracy (5/5) on seeded incident evaluation scenarios.
- Delivered a unified platform combining observability, AI investigation, documentation intelligence, and operational guidance.
What's next for OpsCopilot
We plan to extend OpsCopilot with:
- Native AWS CloudWatch integration
- Incident history retrieval using Retrieval-Augmented Generation (RAG)
- Slack and Microsoft Teams integrations
- Multi-user collaborative incident investigations
- Automated incident timeline generation
- Intelligent anomaly detection
- Role-based access control (RBAC)
- Enterprise authentication
- Multi-cloud observability support (AWS, Azure and GCP)
Built With
- agentic
- fastapi
- geminiapi
- javascript
- python
- react
Log in or sign up for Devpost to join the conversation.