I build the harnesses that let AI actually do things — desktop agents that click buttons, phone calls that translate in real-time, browsers that drive themselves. Applied AI, multi-agent systems, and the infrastructure to make them reliable.
| Project | What it does | Built with |
|---|---|---|
| agent-desktop | Desktop automation CLI for AI agents via OS accessibility trees | Rust |
| cracked-agent | Autonomous browser automation powered by LLMs | TypeScript |
| pilot | Multi-agent autonomous automation framework with 99%+ accuracy via accessibility APIs, computer vision, OCR, and vision LLMs | Python |
| remail | AI email template generator that turns prompts and reference images into production-ready HTML emails | TanStack / Vite |
| soham | Real-time desktop activity tracker with live analytics | Tauri / React / Rust |
| snowden | Export conversations from Claude, ChatGPT and Grok | JavaScript |
| commit-blog | Generate blog posts from git commits using LLMs | TypeScript |
| twilio-realtime-translation | Bidirectional real-time translated phone calls | Python / FastAPI |
| d-id-nextjs | Live AI avatar streaming with voice and GPT-4o | Next.js / TypeScript |
| cursor-cam | Tiny floating camera overlay for macOS that follows your cursor — captured natively by any screen recorder | Swift |
| lighthouse | Fork of Google Lighthouse with a custom Baseline readiness audit — scores pages on browser support coverage with budget thresholds and unresolved token warnings | JavaScript |
lahfir.com · X · LinkedIn · Sponsor