# DossierKit 示例专家

企业级 Agent 平台与大模型应用架构专家

一个虚构示例档案：10 年工程与 AI 产品落地经验，专注将 LLM/Agent 从演示原型落到可评测、可审计、可扩展的企业核心流程。

Location: 中国 / 可远程 / 可 relocation
Experience: 10 years

## Target roles
- Agent 平台架构师: Agent Runtime, Tool-use, LLMOps, Evaluation, Permission & Audit, Enterprise Integration
- 后训练应用算法专家: SFT, DPO, GRPO, RLVR, Data Flywheel, Evaluation Harness
- AI Forward Deployed Engineer: Discovery, Workflow Mapping, Prototype to Production, Customer Integration

## Featured work
- Agent 后训练数据飞轮: 把线上 Agent badcase 转成可评测、可标注、可训练的偏好数据闭环。 (/zh/work/posttraining-data-flywheel)
- 企业级 Agent 平台: 建设可评测、可审计、可扩展的企业级 Agent Runtime 与工具调用体系。 (/zh/work/enterprise-agent-platform)

## Lab
- DPO for Tool-use Preference: 用 chosen/rejected 偏好数据优化企业 Agent 的工具调用决策。 (/zh/lab/dpo-tool-use)
- Agent 回放评测框架: 用生产 trace 构建可复现的 Agent 回归评测，支撑 prompt、工具和模型变更。 (/zh/lab/eval-harness)

## Writing
- 企业 Agent 评测手册: 把 Agent 评测从主观试用变成可回放、可分层、可发布门禁的工程系统。 (/zh/writing/agent-evaluation-playbook)

## Contact
Email: demo@example.com

---

# DossierKit Demo Expert

Agent Platform & LLM Application Architect

A fictional demo profile with 10 years of engineering and AI product delivery experience, focused on turning LLM and Agent prototypes into evaluable, auditable, scalable enterprise workflows.

Location: China / Remote / Open to relocation
Experience: 10 years

## Target roles
- Agent Platform Architect: Agent Runtime, Tool-use, LLMOps, Evaluation, Permission & Audit, Enterprise Integration
- Applied Post-training Algorithm Expert: SFT, DPO, GRPO, RLVR, Data Flywheel, Evaluation Harness
- AI Forward Deployed Engineer: Discovery, Workflow Mapping, Prototype to Production, Customer Integration

## Featured work
- Agent Post-training Data Flywheel: Turned production Agent badcases into evaluable, labelable, trainable preference data loops. (/en/work/posttraining-data-flywheel)
- Enterprise Agent Platform: Built an evaluable, auditable, scalable Agent runtime and tool-use layer for enterprise workflows. (/en/work/enterprise-agent-platform)

## Lab
- DPO for Tool-use Preference: Used chosen/rejected preference data to improve enterprise Agent tool-use decisions. (/en/lab/dpo-tool-use)
- Agent Replay Evaluation Harness: Built reproducible Agent regression evaluation from production traces for prompt, tool, and model changes. (/en/lab/eval-harness)

## Writing
- Enterprise Agent Evaluation Playbook: Turning Agent evaluation from subjective trial into replayable, layered, release-gating engineering. (/en/writing/agent-evaluation-playbook)

## Contact
Email: demo@example.com