Project Description
DreamDefender is an AI-powered autonomous incident response system that detects, analyzes, fixes, tests, and deploys production code changes without human intervention. It uses isolated sandboxes for validation and automated rollback/escalation on failure, reducing time to resolution.
Problem & Impact
Modern apps suffer frequent incidents that require fast response. Manual resolution takes time, causing downtime, revenue loss, and engineer burnout. DreamDefender fully automates detection-to-resolution, achieving:
- lower time to resolution
- near-zero manual debugging
- safe deployments via automated testing + rollback
Technical Architecture
- Backend: FastAPI (Python) – WebSockets, incident simulation, agent orchestration
- Frontend: React (Vite) – real-time UI via WebSockets
- AI Agents: Claude 3.5 Haiku – analysis, reasoning, fix generation
- Sandboxing: Daytona SDK – isolated test environments
- Code Review: CodeRabbit CLI – automated review of AI fixes
- Monitoring: Sentry Webhooks – real-time incident detection
Core Modules:
- Incident detector
- AI reasoning engine
- Code patcher
- Test validator
- Auto-rollback system
- Voice alerting
Pipeline Flow:
- Sentry detects production error → sends webhook to DreamDefender
- FastAPI backend receives incident → orchestrates response pipeline
- Claude 3.5 Haiku (Triage Agent) analyzes root cause and identifies buggy code
- Claude 3.5 Haiku (Fixer Agent) generates code patch
- CodeRabbit CLI reviews AI-generated fix for security/quality issues
- Daytona SDK spins up isolated sandbox (92ms) → applies fix → runs tests
- Auto-deploy on success OR auto-rollback + escalate on failure
- Frontend (React/Vite) displays real-time progress via WebSocket streaming
Tech Stack:
- Backend: FastAPI (Python)
- Frontend: React + Vite
- AI: Claude 3.5 Haiku (2 specialized agents)
- Sandbox: Daytona SDK
- Code Review: CodeRabbit CLI
- Monitoring: Sentry Webhooks
- Real-time: WebSockets + voice alerts
Sponsor Tools & Integrations
This project integrates sponsor technologies as core infrastructure components of the autonomous incident response pipeline:
1. Daytona SDK
Used for secure, isolated sandbox execution of AI-generated fixes.
For every incident, DreamDefender automatically provisions a fresh Daytona workspace (~92ms spin-up), applies the generated patch, and runs the full test suite in a clean environment. This guarantees that validation occurs without any risk to production systems and ensures deterministic, reproducible testing before deployment.
2. Sentry Webhooks
Integrated as the real-time incident detection and monitoring layer.
Sentry sends webhook events to the DreamDefender FastAPI backend whenever exceptions, crashes, or performance anomalies occur in production. These events automatically trigger the autonomous response pipeline, providing structured error data, stack traces, and context to the AI agents for root cause analysis.
3. CodeRabbit CLI
Used for automated AI code review and security validation.
All AI-generated patches are passed through CodeRabbit before sandbox testing. CodeRabbit analyzes changes for:
- security vulnerabilities
- code quality issues
- best-practice violations
- potential regressions
This creates an automated governance layer that prevents unsafe or low-quality code from reaching deployment.
4. Claude 3.5 Haiku (Anthropic)
Powers the autonomous reasoning and code generation system.
DreamDefender uses Claude 3.5 Haiku via streaming API integration with two specialized agents:
- Triage Agent → performs root cause analysis and fault localization
- Fixer Agent → generates targeted code patches and remediation logic
This agent architecture enables fast, cost-efficient, and scalable autonomous incident resolution.
Log in or sign up for Devpost to join the conversation.