Journal
Technical deep dives, lessons learned, and notes from building production AI and backend systems.
A deep dive into SuiteTalk — a multi-tenant AI voice concierge for hotels that routes physical room phone calls via Asterisk PBX to Google Gemini's Multimodal Live API. Learn how we handle raw 8kHz-to-24kHz audio bridging, dynamic system prompt assembly, non-blocking tool calls, and real-time dashboard updates.
How I think about building web scrapers that fix themselves — Claude writes the code once, the scraper runs on autopilot, and when a site changes its layout, Claude analyzes the new HTML and updates the code. Minimal AI, maximum automation.
How I built an MCP server that lets Claude AI autonomously write MQL5 Expert Advisors, compile them, run backtests, scan markets, and analyze trade history — collapsing the entire algo-trading research loop into a single conversation.
A deep dive into every feature of the roman_converter npm package — core conversion, arithmetic, extended numerals up to 3,999,999, classical Latin words, base-12 Roman fractions, a live clock, React hooks, and a full CLI. Zero dependencies, full TypeScript support.
A detailed walkthrough of building CoderSathi — a production-style AI coding assistant with a LangGraph ReAct agent, real-time WebSocket streaming, JWT auth, per-conversation workspaces, and a custom MCP server providing 25+ developer tools.
How I built a TypeScript MCP server exposing 25+ developer tools — file I/O, Git, code analysis, shell execution, and HTTP — for AI coding agents to use via the Model Context Protocol.