Home Page
Connect Repo
Repo Connected
CI Failure- CI AI Fix and PR Creation
Real time status of PR (Merged or Closed)

DevStreamAI — Automated CI/CD Debugging with AI

Inspiration

CI/CD failures are unpredictable, repetitive, and time-consuming. Engineers often spend hours debugging logs, identifying root causes, and applying similar fixes across multiple repositories. We wanted to build an automated DevOps system that not only detects failures but understands them — and then actively fixes them. With GCP Vertex AI, Confluent Cloud, and serverless infrastructure, we saw an opportunity to automate CI debugging end-to-end.

What it does

DevStreamAI listens to CI/CD failure events coming from a Confluent Cloud Kafka topic, analyzes them through an AI engine powered by GCP Vertex AI, and generates human-readable explanations along with patch diffs.

The system then automatically:

Creates GitHub pull requests
Applies AI-generated patches
Notifies developers via Slack or email

A Streamlit dashboard displays CI failure events, explanations, and PR activity in real time.

How we built it

Confluent Cloud (Kafka)

Acts as the central real-time streaming pipeline.
CI pipelines publish failure logs to Kafka, and the consumer triggers the full AI and automation workflow. DevStreamAI uses three Kafka topics in Confluent Cloud:

ci_failures – Streams raw CI/CD failure logs.
ci_pr_updates – Emits updates about PR creation, merge status, and patch operations.
ci_ai_fix – Stores AI-generated explanations and code patches.

GCP Vertex AI

Handles the intelligence layer:

Parsing CI failure logs
Detecting root causes
Generating human-readable explanations
Generating code patch diffs

GCP Firestore

Firestore stores configuration and metadata:

Repository list
Project-to-repository mapping
Dynamic system settings

Firestore makes the system scalable across multiple repositories.

FastAPI Backend on Cloud Run

A containerized FastAPI backend deployed on Cloud Run.

It exposes secure endpoints for:

AI processing
GitHub automation
Dashboard data retrieval

Cloud Run API URL:
https://devstream-backend-176657413002.us-central1.run.app

GitHub Automation Layer

Uses GitHub APIs to:

Create branches
Apply AI-generated patches
Open automated pull requests

Ensures end-to-end CI fix automation.

Streamlit Dashboard

A real-time dashboard that displays:

Failure logs
AI explanations
Patch diffs
PR creation activity

Everything updates live as Kafka events arrive.

GCP Compute Engine VM

A Compute Engine VM hosts:

Kafka consumer
Streamlit dashboard
NGINX reverse proxy

It runs continuously to handle workloads unsuited for Cloud Run, such as:

Long-running Kafka consumers
Real-time dashboard updates
Background processing

Challenges we ran into

Managing sensitive credentials safely using GCP Secret Manager and environment variables.
Coordinating a multi-cloud architecture (GCP + Confluent + GitHub) with consistent retries and ordering.
Making AI-generated patches accurate and maintainable across diverse repository structures.

Accomplishments that we're proud of

Built a fully automated AI-driven DevOps pipeline.
Achieved end-to-end flow: CI failure → Confluent Kafka → AI analysis → patch → GitHub PR.
Designed a scalable multi-repository architecture powered by Firestore.
Built a real-time monitoring dashboard.
Significantly reduced debugging time for repetitive CI failures.