Inframind

• 

  authentication page

• 

  tfstat file is sync automatically with any changes

• 

  destroy resource with showing cost estimation, security issues, rollback strategy, tfstate uploaded

• 

  create resource with showing cost estimation, security issues, rollback strategy, tfstate uploaded

Inspiration

The Problem: Cloud infrastructure is a "walled garden." You either need to be a Terraform expert or spend hours in the GCP Console clicking buttons.

We were inspired by the gap between design and deployment. Why can't we just tell a machine what we want—or show it a sketch—and have the infrastructure appear? Inframind was born to turn "Architectural Intent" into "Running Resources" instantly.

What it does

Inframind is an AI-powered SRE Agent that acts as a bridge between your ideas and Google Cloud.

Prompt-to-Provision: Tell the agent to "Deploy a GKE cluster with a specific VPC," and it writes the Terraform code for you.

One-Click Apply: It doesn't just write code; it initializes, plans, and applies the infrastructure directly.

Infrastructure Brain: It manages your GCP state, handles errors, and remembers your configuration.

Visual-to-Code (Beta): We are currently refining the engine to allow users to upload architectural diagrams and have them converted into code.

How we built it

Core AI: Gemini 3.1 Pro Preview for high-reasoning HCL generation.

Execution Engine: Custom Python backend using subprocess to trigger Terraform CLI.

Frontend: Streamlit for a real-time, dark-mode terminal experience.

Deployment: Fully containerized with Docker for environment consistency.

Challenges we ran into

Real-time Feedback: Streaming Terraform logs into a web UI while the process is running in the background was a major threading challenge.

State Locking: Handling situations where the AI might try to deploy something while a previous job is still locking the state.

Accomplishments that we're proud of

The "Loop": Creating a system that successfully runs terraform plan, presents it to the user, and waits for a human "OK" before applying.

Zero to Cloud: Successfully deploying complex VPC and GKE setups from a single chat message.

What we learned

We learned that Gemini 3.1 is remarkably good at "self-correction." If the Terraform plan fails, the agent can read the error logs and rewrite its own code to fix the issue without human intervention.

What's next for Inframind

Finish Diagram-to-Code: Perfectly mapping visual nodes (load balancers, DBs) to specific GCP modules.

Autonomous Monitoring: Agents that watch your logs and automatically suggest infrastructure changes if traffic spikes.

DevSecOps Agent: A specialized agent that "pre-audits" code for security flaws before the apply phase.

Built With

• gcp
• gemini
• hcl
• python
• terraform