Wang Jianjun

About

From data systems to agents: products people can use, inspect and improve.

China · UTC+8Tencent · Backend DevelopmentSZTU · Data Science

I am Wang Jianjun. I work on data systems and AI applications, starting with concrete tasks and making tool use, memory and delivery easier to inspect and trace.

At Tencent, I contribute to Go monitoring services, focusing on ingestion backpressure, statistical consistency and cross-day metering. Previously, I worked on design-agent evaluation and tooling at Kuaishou.

I start with the task a person needs to complete, then decide which steps belong to a model, a rule or a human reviewer. LabMemory uses human confirmation; Shared Memory Bus keeps source records. Both follow that approach.

Capabilities in practice

Connecting skills to projects

Select a project
WJENGINEERINGEvaluationTrusted memoryMCP servicesProduct interfacesRetrieval & checksCode intelligence
Practice in a project

Evaluation

Check delivery validity, evidence coverage and requirement fidelity for editable design artifacts, then repair findings and rerun the same case.

Agent Delivery & EvalsFigma · Evidence · RegressionRead the project

Engineering Principles

Core Beliefs

Business-First Delivery Loop

Business-First Delivery

The ultimate success metric of an AI system is not generation volume, but whether it reliably reduces costs and amplifies throughput in real operations.

Strict Tool Contracts & Governance

Strict Tool Contracts & Governance

Tool invocations must enforce strict schema validation, deterministic structural outputs, and clear circuit breakers to prevent cascading hallucination.

Triple-Layer Verifiable Evaluations

Triple-Layer Verifiable Evals

A single HTTP 200 or non-empty string is not acceptance. We combine native structural validation, rendered visual proof, and business requirement audits.

Obsessive Traceability & Regressions

Obsessive Traceability & Regressions

Freezing prompt context, decision traces, intermediate artifacts, and error snapshots ensures sporadic failures can be diagnosed and regression-tested.

Capability map

Skills & Mastery

AI Agent Systems

Designing and building agentic systems that reason, retrieve, edit and act.

  • MCP
  • Agent Evaluation
  • Agent Harness
  • LangGraph
  • Multi-Agent
  • Trace / Replay

Retrieval & Evals

Building verifiable benchmarks, hybrid retrieval and data flywheels.

  • RAG / Hybrid Retrieval
  • LLM-as-a-Judge
  • Context Engineering
  • Tree-sitter / AST
  • Agent Observability

Engineering & Backend

Shipping reliable, concurrent backend systems and toolchains.

  • Python
  • TypeScript
  • FastAPI
  • React / Node.js
  • Redis
  • asyncio
  • Docker
  • Git

Product & Delivery

Turning complex requirements into verifiable, regressible production capability.

  • PRD-driven
  • Figma Structured Editing
  • Data Flywheel
  • Business Delivery

Honors & Competitions

Achievements