Projects

Specialist agents from an enterprise multi-agent platform for plant operations. Each page is a one-page writeup. Official demos will be linked as they come online.

Multi-Agent Orchestrator

Conversational entry point that classifies user intent, routes to specialist agents, and orchestrates multi-step autonomous plans with human approval.

LangGraphMulti-AgentPython
Enterprise project →

Voice Control Bridge

Hands-free, real-time audio control of the plant system with native speech understanding, barge-in, and natural turn-taking.

Realtime AudioWebSocketReact
Enterprise project →

Root Cause Analysis Agent

A parallel agent swarm that investigates equipment failures across an institutional knowledge graph, the public web, and internal documentation, with human-in-the-loop plan review.

LangGraphAgent SwarmRAG
Enterprise project →

Data Analysis Agent

Generates and runs Python in a secure sandbox to answer statistical, ML, and visualization questions about instrument time-series data.

LangGraphCode SandboxPython
Enterprise project →

Document Intelligence Agent

Multi-modal document Q&A, structured data extraction from specification PDFs, and on-demand technical image generation, all routed through a single decision graph.

LangGraphMulti-modalStructured Output
Enterprise project →

Autonomous Task Executor

Plans, approves, schedules, and runs long-running workflows — with cross-agent delegation, anomaly detection, chart-backed reports, and multi-step chains.

LangGraphSchedulingWorkflow Automation
Enterprise project →

GUI Automation Agent

Controls the plant dashboard by reasoning over screenshots, with fingerprint-based element remapping and dynamic schema constraints that keep the model in bounds.

LangGraphVisionSet-of-MarksReact
Enterprise project →

Scientist-Bin

Agentic data scientist that plans ML/DL experiments, writes the notebook, and reports back — a one-person research pipeline in chat form.

PythonAgenticMLDeep Learning
Open source →

I Can Do All Things

A from-scratch study trail through deep learning — built to learn by implementing, not by reading.

Deep LearningNotebooksSelf-study
Open source →

CodeSense

Drive Claude Code with hardware you already own — analog-trigger approvals, lightbar agent status, haptics, a touchscreen orb.

TypeScriptClaude CodeHardwareHID
Open source →

Infinite Canvas Agent

An infinite engineering canvas where agents read and draw P&IDs — equipment as nodes, piping as edges.

LangGraphVue 3GeminiP&ID
Open source →
© 2026 Dr. Bin Liu