Rocky
A general-purpose agent console. Multi-model routing, tool use, and a clean conversational interface — built as a foundation for deploying custom AI agents.
Things I've built, mostly with LLMs.
A general-purpose agent console. Multi-model routing, tool use, and a clean conversational interface — built as a foundation for deploying custom AI agents.
Built on Rocky for a small e-commerce business. Three workflows: a data analysis agent for sales and reporting, an automated customer service agent that searches historical cases and drafts replies, and an inventory management agent that combines deterministic forecasts with model review.
A multi-agent system that researches, writes, and produces executive briefing PDFs. Thirteen phases from web research through writing, editing, and vision-based PDF polishing. View a sample issue →
Real-time voice AI for practicing consulting case interviews. LiveKit for audio, Deepgram for transcription, ElevenLabs for speech — an interactive interviewer that adapts in real time and provides structured performance feedback.
Hierarchical LLM-powered workforce scheduling. An orchestrator delegates to sub-agents for roster construction and cost optimization, balancing department assignments, leave, shift conflicts, and labor costs. Written about in Building an Agent for Scheduling.
My first attempt at building something with LLMs. An early experiment in chaining "agents" — which had no tools, just specialized prompts — with recursion and looping. Clearly outdated now, but a look at the early shift from chatbots to agent traces.