Our Tech Stack for Speed, Cost, and Scale

How We Built Vortex IQ to Be Fast, Lean, and Future-Ready

Startups live and die by three levers: Speed, Cost, and Scale.

At Vortex IQ, we’re building an AI agentic platform that transforms e-commerce APIs into intelligent agents. From staging environments to SEO automation, our users rely on us to move fast, stay affordable, and scale predictably.

That meant choosing a tech stack that matched our ambition, without overengineering or overspending.

Here’s a transparent look at how we built our stack to deliver on all three fronts.

Speed: From Prompt to Production in Minutes

Frontend

• Framework: React with Vite
• Styling: Tailwind CSS
• UI Components: shadcn/ui, Radix UI
• Bundling: Vite → Blazing fast HMR and build times
• Realtime: WebSockets for live agent updates

***Why it works*:** Rapid iteration. Reusable components. Sub-50ms UX on mission-critical views like dashboards, logs, and prompt interfaces.

Backend

• Framework: Laravel (PHP) + Octane
• Worker Layer: Laravel Horizon (queue-based)
• Web Server: RoadRunner or Swoole for concurrency
• Auth: Laravel Sanctum (API token-based)
• Rate Limiting: Custom throttling at agent and endpoint levels

***Why it works*:** Battle-tested, clean architecture. Great dev velocity + async processing out of the box.

Prompt/Agent Layer

• LLMs: OpenAI (GPT-4o), Claude 3, Llama 3 (Meta)
• Routing Logic: Custom Model Context Protocol (MCP)
• Execution Engine: Multi-agent orchestration with fallback, retry, reasoning chain
• Logging: Full visibility of agent state, actions, errors, and outcomes

***Why it works*:** Every prompt is routed through our MCP layer for context-rich, safe execution across APIs.

Cost: Optimising for Burn, Not Just Brilliance

Cloud Infrastructure

• Primary Hosting: Vultr (for cost/performance edge)
• Container Orchestration: Docker + Docker Compose (simple, no Kubernetes overhead)
• CI/CD: GitHub Actions + Laravel Envoy
• Image Optimisation: Squoosh CLI + AVIF + WebP pipelines
• Database: PostgreSQL with pgbouncer pooling

***Why it works*:** Every component was selected to avoid vendor lock-in and reduce compute/storage waste.

Model Cost Control

• Prompt-level cost monitoring
• Fallback to open-source models (LLaMA, Mistral) when task complexity allows
• Token caching and deduplication to prevent redundant calls

***Why it works*:** LLMs can eat your margin. Our routing logic balances performance with cost efficiency.

Scale: Designed to Handle 10x Without Rewrites

Modularity

• All AI agents are microservices with individual repos, scopes, and tests
• Agents register themselves with the Agent Registry and auto-sync to our UI
• Internal CLI tools to scaffold new agents in minutes

API Layer

• JSON Schema-based validation for every incoming/outgoing API interaction
• API Gateway abstraction to support Shopify, BigCommerce, Adobe Commerce, Google Analytics, etc.

Observability & Rollback

• Centralised logging with Papertrail
• Real-time alerts for agent anomalies
• Rollback system for every action (e.g. undo a price change, revert SEO update)

***Why it works*:** We scale horizontally. Agents are disposable and restartable. Nothing is hardcoded.

Final Thought

In a world racing towards agent-led systems, the stack is not just tech, it’s strategy.

We didn’t chase the shiniest tools. We optimised for:

• Developer velocity
• Infrastructure cost
• Future-proof execution
• Real-time performance

This is the architecture that helps us ship weekly, serve global merchants, and stay capital-efficient, even as we scale to 3,000+ retailers.

Want to know how our agentic platform could power your use case? Visit vortexiq.ai or write to sambit@vortexiq.ai.