AI Infrastructure & Distributed Systems

Quanta – AI Browser Execution Engine

Role: Founder / Principal Software Engineer   //   Timeline: Dec 2024 – Present

Quanta – AI Browser Execution Engine architecture

Core Metrics

  • Sustained 10 concurrent Playwright contexts within a strict 2GB container budget
  • Enforced hard caps of 40,000-token execution limits to prevent runaway LLM loop costs
  • Bypassed WAFs successfully using AES-256-GCM encrypted browser-session storage

Technology Stack

Languages

Go, Python, TypeScript

Orchestration

Temporal, NATS JetStream, Docker

Data Layer

PostgreSQL, Redis, Qdrant, MinIO

Frontend

Next.js, Playwright

The Problem & Constraint

An AI-powered web automation platform built to execute complex, authenticated browser workflows securely. Bypasses expensive multimodal vision models in favor of a fast, deterministic DOM-first extraction engine.

System Topology

The system is strictly decoupled into a Go-based Control Plane (handling REST, JWT validation, and atomic billing) and a Python Execution Plane. Communication is orchestrated via gRPC and Temporal workflows. Live execution telemetry is buffered in memory via NATS JetStream and piped to the frontend via Server-Sent Events (SSE) to ensure the Python worker is never blocked.

Engineering Tradeoffs

  • 01

    Engineered a deterministic JavaScript schema-walker and structural DOM fingerprinting to extract data directly, invoking Llama-3.1-8B and Gemini 2.0 Flash purely as dynamic fallbacks.

  • 02

    Implemented semantic RAG caching using Qdrant and all-MiniLM-L6-v2 embeddings to match new navigation objectives with structurally similar, previously successful execution sequences.