Selected work.
Documented results across AI infrastructure, automation, and platform delivery.
Private LLM Gateway with Local Compute
Designed and deployed a local, high-availability LLM gateway utilizing vLLM and LiteLLM proxy on dedicated GPU arrays. Serving local open-source models like Qwen 27B, the setup provides unlimited, low-latency inference for parallel agentic workflows while maintaining strict data privacy compliance and cutting token API overhead by up to 80%.
High-Density WhatsApp Automation
Replaced resource-heavy headless browser scraping nodes (Playwright/Puppeteer running at 2.5GB RAM per instance) with a custom Go-based WebSocket integration using Whatsmeow. This minimized the memory footprint to just 100MB per client node—a 96% reduction—allowing horizontal scalability of over 60 concurrent automation nodes on standard hardware.
Secure SSH automation for AI agents
Go-based MCP server that delegates SSH via local profiles and ssh-agent, keeping credentials off the network. Append-only audit logs trace every LLM-assisted infrastructure action.
Automated On-Premise Runner Fleet
Designed and automated a containerized, self-hosted GitHub Runner fleet running on local infrastructure. This decoupled heavy integration testing suites from public runner queues, providing unlimited build minutes and sub-minute feedback loops for staging-to-production pipelines.