DevWatch AI

• 

  System app to register products

• 

  Live App to track production

• 

  Auto PR creation based on production error logs

• 

  Digital Oceal - Knowledge Base sync - Github action

• 

  System app to track events, PR and function calls

Inspiration

Production failures don’t usually happen because engineers don’t care — they happen because follow-ups get delayed.

We were inspired by real-world incidents like large-scale AWS service outages, Cloudflare edge failures, and react2shell-style supply-chain vulnerabilities, where early signals existed but fixes took hours or days to land. Logs were noisy, vulnerability feeds were overwhelming, and translating an incident into a safe code change was still a manual process.

What if production incidents automatically turned into pull requests?

---

What it does

DevWatch AI continuously monitors production systems for runtime errors, health regressions, and dependency vulnerabilities. When an issue is detected, it:

1. Analyzes the root cause using DigitalOcean Gradient AI, grounded in the project’s codebase
   
2. Generates a fix using OpenAI GPT-OSS 120B via Gradient Serverless Inference
   
3. Opens a detailed GitHub pull request, with optional auto-merge when safety rules are met
   

This turns incidents into fixes in minutes instead of hours.

---

How we built it

DevWatch AI runs entirely on DigitalOcean:

• DO Functions run scheduled monitors for logs, health checks, and CVEs
  
• Logs and artifacts are stored in DigitalOcean Spaces
  
• Code is synced into a Gradient Knowledge Base via GitHub Actions
  
• A background Event Processor calls:
  • Gradient Agent API for root-cause analysis and fix strategy
    
  • Gradient Serverless Inference (GPT-OSS 120B) for code generation
    
• Fixes, events, and audits are stored in DO Managed PostgreSQL
  
• PRs are created and optionally auto-merged via the GitHub API
  

A real-time Next.js dashboard shows system health, events, and generated fixes.

---

Challenges we faced

• Preventing unsafe auto-merges required confidence scoring, diff limits, and restricted file rules
  
• Raw logs were too large for AI analysis, so we built structured event extraction
  
• Keeping fixes deterministic for live demos required controlled failure scenarios
  
• Designing AI workflows that assist — not replace — human review
  

---

What we learned

• AI is most effective when constrained by guardrails
  
• Knowledge-base grounding dramatically improves reliability
  
• Automation should optimize incident follow-ups, not bypass engineers
  
• Production tooling must be explainable to earn trust
  

---

What’s next

• Multi-language support (Python, Java, Go)
  
• Distributed tracing integration
  
• Multi-repo monitoring
  
• Slack and CI/CD integrations
  
• Advanced incident analytics
  

---

Built With

• Languages: JavaScript, TypeScript
  
• Frameworks: Node.js, Next.js
  
• Cloud Platform: DigitalOcean
  
• AI: DigitalOcean Gradient AI Platform, OpenAI GPT-OSS 120B
  
• Databases: DigitalOcean Managed PostgreSQL
  
• Storage: DigitalOcean Spaces
  
• APIs: GitHub API, NVD, npm advisories
  

---

Try it out

• GitHub Repository: https://github.com/your-username/devwatch-ai
  
• Live Dashboard (Demo): http://localhost:3000
  
• Demo App: http://localhost:3001
  

---

Video Demo

Add your demo video link here.

Built With

• digitalocean
• do-agent
• do-function
• do-knowledge-base
• do-manged-db
• do-serverless-inference
• github
• github-functions
• gradient-ai
• nextjs
• postgresql
• worklows