Agent Harness
Task lifecycle, context selection, controlled tools, artifacts, snapshots, verification, and replay.
Technical Profile / Extended Reading
Engineering practice for complex relationship data and Agent Harness
I work across graph application engineering, financial data systems, GraphRAG, and AI Agent Harness design. This site adds context around my work, capability structure, and AI engineering methods.
The focus is not a repeated resume. It shows how real systems handle task state, context selection, tool boundaries, evidence chains, artifacts, verification, and replay.
01
Task Lifecycle
Sessions, tasks, states, logs, and artifacts are modeled explicitly.
02
Tool Boundary
The model proposes operations; the system validates, executes, and records them.
03
Replay Basis
Real tasks become material for diagnosis, replay, and continued iteration.
Capability Structure
Putting model capability inside real systems
These capabilities come from long-running projects including GraphPilot, relation-graph, Hovo, GraphRAG, Guanlan, and local A-share quantitative research.
Graph application engineering
Relationship data modeling, layout, interaction, editing, multi-framework adaptation, and graph knowledge assets.
Knowledge engineering
Document trees, entity relations, vector recall, graph expansion, rerank, and fact-constrained answers.
Financial data systems
Banking semantics, metric definitions, evidence chains, report generation, and local quantitative replay.
Representative Work
Technical judgment through system architecture
Contact