DevPulse Studio

Autonomous Multi-Agent AI System powered by Gemini & ClickHouse for Real-Time Self-Healing Media & VFX Render Pipelines.

Comment

Inspiration

Modern film studios and streaming platforms process millions of high-resolution video frames daily using complex media pipelines (FFmpeg, render farms, edge proxies). When a worker crashes due to Out-Of-Memory (OOM), missing CUDA drivers, or gateway buffer limits, engineers spend hours reading raw logs, manually diagnosing issues, and redeploying fixes.

DevPulse Studio was created to revolutionize incident management in cinema engineering by building an Autonomous, Self-Healing Multi-Agent AI Platform that reduces Mean Time to Resolution (MTTR) from hours to seconds.

What it does

DevPulse Studio automatically ingests live crash telemetry from media render nodes and orchestrates a team of specialized AI Agents to triage, resolve, verify, and govern pipeline failures.

  • 🔍 Autonomous Triage Agent: Parses raw stderr logs, exit codes, and hardware telemetry to instantly identify root causes.
  • 🤖 AI Fix Generator Agent: Generates precise configuration/code diff proposals (e.g. dynamic buffer pool limits, memory scaling).
  • 🧪 Verification Agent: Validates proposed patches in an isolated containerized sandbox before pushing to production.
  • 🛡️ Governance & Rollback Guardrails: Evaluates outage severity and automatically triggers emergency guardrails or requests human-in-the-loop sign-off.
  • 📊 Real-Time ClickHouse Analytics: Provides high-throughput dashboard analytics tracking MTTR, ROI (engineer hours saved), error frequency, and active worker statuses.

How we built it

We designed DevPulse Studio from the ground up with high performance and enterprise reliability:

  • Frontend: Built using Next.js 16 (App Router, Turbopack) and React 19, styled with a custom SmartPG SaaS Dark Theme (#0B0E14 deep navy canvas, #121824 cards, and #1F293D crisp borders).
  • AI & Agentic Layer: Powered by Google Gemini 1.5 Flash orchestrated via Python. The agents utilize the Model Context Protocol (MCP) to interact with ClickHouse safely via schema-validated tooling.
  • Data Engine: ClickHouse Cloud serves as the unified event log, handling high-concurrency crash ingestion, incident event timelines, and materialized views for sub-second dashboard rendering.
  • Deployment: Hosted on Vercel with seamless API route handlers for ingestion and streaming telemetry.

Challenges we ran into

  1. Timezone & Real-Time Syncing: Normalizing ClickHouse UTC timestamps with client browser local times so incident cards display exact real-time tags (just now, 5s ago) without offset drift.
  2. Strict MCP Protocol Adherence: Ensuring AI agents operate exclusively through schema-validated Model Context Protocol (MCP) tool interfaces rather than raw arbitrary SQL injections.
  3. Emergency Guardrail Logic: Implementing a reliable human-in-the-loop approval workflow for high-risk critical outages while maintaining auto-resolution for low-risk patches.

Accomplishments that we're proud of

  • ⏱️ Achieved an automated MTTR under 10 seconds for standard render node failure patterns.
  • 🛡️ Successfully built a reliable Human-in-the-Loop Governance Guardrail that safely halts critical production deployments until human sign-off.
  • ⚡ Seamlessly connected ClickHouse Cloud + Google Gemini 1.5 Flash via MCP Tools for ultra-fast, structured multi-agent execution.
  • 🎨 Built a stunning, responsive, enterprise-grade dark dashboard (SmartPG Dark Navy Theme) optimized for high-density telemetry viewing.

What we learned

  • How to leverage Google Gemini 1.5 Flash for rapid, accurate multi-agent reasoning over complex technical logs.
  • Designing sub-second analytics dashboards with ClickHouse materialized views.
  • Standardizing AI-driven code modification and verification loops safely in production environments.

What's next for DevPulse Studio

  • 🔮 Predictive Failure Prevention: Using Gemini to analyze telemetry trends and predict OOM errors before nodes crash.
  • 💬 Slack & PagerDuty Deep Integration: Interactive approval buttons directly inside team communication channels.
  • ☁️ Multi-Cloud Render Farm Support: Expanding auto-scaling integrations across AWS Thinkbox Deadline, Google Cloud Batch, and Kubernetes.

Built With

Share this project:

Updates

Submission history