Beyond the Model
Models keep improving, but the real challenge is putting them into real tasks: how to structure context, invoke tools, preserve state, and recover from failures.
INDUSTRIAL AI PRODUCT PORTFOLIO
LIU SHIXIN
Industrial AI Product Lead | Enterprise Agent Applications | Agentic Workflow | Energy & Data Center AI
Turning complex industrial workflows into deployable, auditable, and measurable AI applications.
15 years of complex industry operations experience, combined with hands-on AI product prototyping and enterprise Agent Harness practice.
I focus on how AI systems can be embedded into real business workflows with human review, audit trails, evaluation metrics, cost visibility, and operational feedback loops.
ContractAI is a desktop prototype. The Web Demo uses anonymized samples and simulated parsing results.
15 yrs
Industry Operations
2 yrs
AI Product Practice
10+
AI Prototypes / Blueprints
100M+
Tokens Used (30d)
1M+
Agent Tutorial Views
ContractAI
Desktop Prototype
Figures are for portfolio illustration. Exact data available on request.
Problems I Care About
The hard part isn't the demo. It's getting AI into real workflows — with boundaries, traceability, failure handling, and continuous evaluation.
Enterprise Boundaries
High-risk actions — contracts, payments, approvals, external communications — can't be delegated to AI alone. Products must design for human review, audit trails, replay, and accountability.
Measurable Agent Quality
I don't just ask 'did the demo succeed?' I track task success rate, human takeover rate, failure taxonomy, token cost, and whether feedback is fed back into the evaluation set.
Main Project
ContractAI — Contract Intelligence Agent Harness
Transforms PDF / image contracts into verifiable, correctable, exportable, and traceable structured contract ledgers. Validates human-in-the-loop review, audit trails, and secure local storage in high-stakes enterprise environments.
ContractAI Demo Video
Demo video loading. This area serves as a placeholder when the file is unavailable.
Demo flow: Upload / Parse / Review / Ledger / Exportpreload=metadata
Core Competencies: Bridging Industry Expertise and AI Systems
Complex Business Abstraction
Business Flows / Approval Flows / Contract Flows / Financial Flows / Risk Controls
Agent Harness Productization
Agent Loop / Tool Use / MCP / Memory / Context
Enterprise AI Deployment
Permissions / Human Review / Audit / Acceptance / Risk Boundaries
AI-Native Prototyping
Cursor / Codex / Claude Code / v0 / Bolt / Gemini / Dify / LangGraph / n8n / Coze / Qwen / Ollama
Evaluation & Feedback Loops
Task Success Rate / Failure Taxonomy / Human Takeover Rate / Token Cost
End Goal
Embed model capabilities into real enterprise systems — delivering AI outputs that are confirmable, traceable, measurable, and shippable.
Agent Harness Architecture
Designing AI systems like data pipelines — wrapping uncertain model inference inside stable engineering chains.
More Prototypes & Engineering Explorations
TokenRadar — AI API Cost & Observability Platform
A local LLM API gateway and Token Observability dashboard for tracking token consumption, API costs, failure rates, latency, and anomalous usage across agents, projects, models, and API keys. Extensible into an enterprise AI FinOps & API Key Governance platform.
Local API GatewayToken ObservabilityAI FinOpsAPI Key Governance
OpenClaw / Hermes — Mobile Agent Deployment Research
Deployed open-source agent frameworks on Android devices to validate low-cost local agent operation, environment compatibility, background persistence, and accessibility for general users. Collected real-world feedback through public video tutorials.
OpenClawHermesTermuxLocal AgentUser Feedback
Scan to view public video cases (WeChat)
AI FDE Workbench — Field Deployment Engineer Copilot
Designed for enterprise AI field deployment engineers: a natural-language-driven workbench that assists with customer requirement intake, data ingestion, business ontology modeling, agent assembly, debugging, and delivery report generation.
FDEAgent AssemblyOntologyPlannerEval / Release
Enterprise Knowledge Base & Agentic Workflow
Designed a RAG-augmented enterprise knowledge base and extended it from single-turn Q&A to multi-step task execution, supporting knowledge retrieval, document generation, process progression, human approval, and result archiving.
RAGWorkflowHuman-in-the-loopPermissionAudit Log
Career Path — From Industry Operations to AI Product
From real estate operations, engineering management, and clean energy investment, to AI-native product practice.
2023–至今 / Present
Independent AI Product Lab / Enterprise AI Deployment
Agent systems, RAG, OCR, AI workflows, TokenRadar, ContractAI, FDE Workbench
2023–至今 / Present
Co-founder, C&I Distributed Solar Investment Company
Project acquisition, financial modeling, contract coordination, operations management, industrial AI validation
2021–2022
Shidi Group
Group operations management, 159 active construction projects, risk governance, subsidiary coordination
2019–2021
Kaisa Group
Large-scale operations, investment, financing, management reporting, budget management
2009–2019
Longfor Group
Complex project management, engineering operations, organizational development, process standardization