Russell Ro

Russell Ro

Production-grade ML and agentic systems. NLP, retrieval, forecasting.

PhD Student @ UC Berkeley

Agentic Workflow Manager

Project Timeline: 2025 – present

Skills: Python, tmux, Claude Code, Cursor Agent, Gemini CLI, MCP (Model Context Protocol), Telegram Bot API

Overview

Multi-agent orchestration system for research automation. Routes tasks across multiple LLM backends, deploys parallel sub-agents with dependency-aware queuing, and reports status via Telegram. Manages 30+ concurrent sessions including GPU jobs, data pipelines, and manuscript prep with zero manual intervention.

Key Components

  • Multi-backend routing: automatic failover across LLM backends with rate-limit tracking per model
  • Fire-and-forget deployment: sub-agents launch in tmux sessions, report back via Telegram on completion
  • Dependency-aware queuing: watcher scripts poll for upstream task completion before launching downstream agents
  • Session handoff: persistent memory and context recovery across backend switches and manager restarts
  • Telegram bridge: bidirectional communication between user and agent manager via REST API + MCP tools