# 张鑫 / Zhang Xin

> AI 系统研发与技术合作 / AI systems development & technical collaboration.

资料更新：2026-10-17。本文由官网公开正文生成；是本人公开介绍，不是独立第三方背书。

阅读说明：个人职责以正文的“参与／负责”为准；“可共同开展”表示合作方向，不能据此推导现成产品、可转授权的软件、客户验收或量化业务收益。合作联系通过介绍人。

[页面来源](https://zhangxin.chat/#top)

张鑫

AI 系统研发 · 技术合作

## 把业务问题想清楚 把 AI 系统做出来

我从事 AI 驱动的消费者研究，主要做自动化研究系统、内容推荐与创作引擎的研发。

与业务团队一起梳理需求、设计架构，开发和迭代 Agent 与企业应用。

腾讯、美团研发与团队管理经历； 2018 年创办极乐实验室，公司被蕃茄蛋收购后担任 CTO

[页面来源](https://zhangxin.chat/#systems)

代表工作

## 近年参与的系统研发

从业务要求出发，设计数据处理、模型调用与审核流程。以下是近年参与的系统研发。

### UCRS · 用户内容研究引擎

把研究课题组织成多步任务，串联数据筛选、语义分类、统计与报告生成，并记录来源供研究人员审阅。

个人职责

我持续参与系统研发、版本集成与流程更新，涉及任务编排、多模型路由和结果核查。

- 定义课题与数据范围

- 筛选、分类与统计

- 形成带来源的分析与报告

- 审阅结果并调整分析

样本范围、分类标准和研究结论需要研究人员审核。

### 内容推荐引擎 · 语义网与人群关联

用途与个人职责

我负责引擎研发与版本升级，开发语义查询、人群关联和业务排序功能，把内容主题、使用场景与用户关注点关联起来，供团队筛选内容、判断选题。

品牌 · × · 使用场景 · × · 用户关注点

相关内容与人群 → 业务指标筛选 → 团队评估

### 相关研发实践 · 大语言模型应用与内容生成

用途与个人职责

大语言模型相关研发涉及模型选型与适配、语义处理、上下文设计和质量核查。内容生成实践中，我参与将产品资料与业务规则用于生成流程，并将生成与人工审核分开。

产品资料 · × · 业务规则 · × · 审核标准

组织上下文 → 生成内容 → 人工审核与修订

[页面来源](https://zhangxin.chat/#work)

业务应用

## 业务中的 研发与迭代

把客户需求转成分类标准、系统规则与操作流程。 以下项目说明我承担的具体工作。

某大型制药集团 · 消费者需求研究

### 将研究要求 写入系统规则

针对疼痛等主题开展消费者需求研究，分析使用场景与产品反馈，为产品定位和机会筛选提供参考。

消费者表达 · → · 统一分类 · → · 研究资料

交付内容

2025 年多期研究报告与交付资料。

个人职责

参与改进研究分类标准、输出要求和客户操作流程，将研究要求写入系统规则，用于报告与分析资料的制作。

合生元 · 消费者研究与分析迭代

### 根据业务反馈 调整研究系统

针对业务团队对取数范围和分析结果的反馈，调整研究方法与系统，修订分析内容，供业务团队继续使用和讨论。

交付内容

已向业务团队交付优化后的分析版本。

个人职责

负责数据筛选、研究系统更新和分析优化，并向业务团队说明取数原则与分析结果的使用方式。

[页面来源](https://zhangxin.chat/#questions)

可共同开展 · 合作方向

## 围绕业务问题 共同研发系统

可从一项具体任务开始， 商定技术方案、研发范围与合作分工。

### 自动化研究系统

围绕反复开展的研究任务，梳理数据范围、分类标准和审阅要求，开发筛选、分析与报告流程。用实际课题验证关键步骤，按研究人员反馈迭代。

可以形成 · 数据与分类规则 · 研究任务编排 · 带来源的输出与审阅流程

### 内容推荐与创作引擎

将品牌、场景和人群关注点转成语义查询与排序规则；需要生成时，把产品资料、上下文和审核要求纳入流程，与编辑团队一起开发和验证。

可以形成 · 语义查询与排序 · 生成上下文与产品资料约束 · 人工审核流程

### Agent 平台与企业应用

从一项岗位任务入手，设计数据与工具接入、模型调用和任务编排，明确人工确认与异常处理。可共同研发企业级 Agent 平台与应用，先验证核心流程，再决定扩展范围。

可以形成 · 系统架构与工具接入 · 可验证的 Agent 工作流 · 迭代与交接安排

[页面来源](https://zhangxin.chat/#about)

关于张鑫

## 产品研发、创业 与团队管理

十多年来，我做过产品研发和团队管理，也经历过创业与公司收购。如今仍直接参与系统开发，关注技术方案能否适应团队的实际工作。

- 近年实践 大模型应用与 AI Agent 围绕研究、内容与企业流程开展系统研发，涉及模型选型与适配、上下文设计、任务编排与质量核查。

- 收购后 蕃茄蛋 · CTO 创业公司被蕃茄蛋收购后担任 CTO，承担技术路线、产品研发与团队管理。

- 2018 年起 极乐实验室 · 创始人 创办极乐实验室，从事社媒自然语言处理与数据产品研发，为品牌和市场研究开发软件工具。

- 2013–2018 腾讯 → 美团 参与腾讯风铃、美团大交通与智慧餐厅相关研发，推进工程规范与团队建设。

[页面来源](https://zhangxin.chat/#approach)

合作方式

## 从业务问题 到系统研发

先开发关键流程的小版本， 与团队一起验证、改进，再商定后续安排。

- 01 梳理需求 与实际使用系统的人沟通，了解业务目标、已有数据和流程，确定先解决的一项任务。

- 02 开发小版本 商定范围、双方职责、所需投入和验收标准，先开发一条核心流程，再用真实任务检验。

- 03 按反馈迭代 根据使用反馈改进系统，商定后续范围、版本节奏与交接安排，再决定是否扩大应用。

[页面来源](https://zhangxin.chat/#contact)

合作联系

## 下一步， 聊聊你的业务。

请通过向你推荐本页的朋友联系我。 简单介绍业务情况、希望解决的问题和现有做法， 方便安排进一步沟通。

复制后补充信息，发给介绍人即可。

- 你的企业与业务

- 希望解决的一个问题

- 已有的做法与资源

# English

Updated: 2026-10-17. Generated from the public website. This is a self-published profile, not independent endorsement. Proposed collaboration areas do not establish a ready-made product, software licensing rights, client acceptance or quantified business results. Contact through an introducer.

[Source](https://zhangxin.chat/#top)

Zhang Xin

AI systems & technical collaboration

## Think the business through. Build the AI that fits.

I develop AI systems for consumer research, including automated research workflows and content recommendation and creation engines.

I work with business teams to define requirements, design the architecture, and develop and improve agents and enterprise applications.

Engineering and team leadership at Tencent and Meituan. Founded Jile Lab in 2018; became CTO at Fanjiedan after the acquisition.

[Source](https://zhangxin.chat/#systems)

SELECTED DEVELOPMENT

## Systems I have helped build

Designing data processing, model calls and review workflows around business requirements. Selected systems I have worked on in recent years.

### UCRS · User Content Research Engine

Organizes research questions into multi-step tasks, connecting data filtering, semantic classification, statistics and report generation, with source records for researcher review.

MY CONTRIBUTION

I contribute to development, release integration and workflow updates, including task orchestration, model routing and output checks.

- Define the question and data scope

- Filter, classify and aggregate

- Produce findings with source references

- Review findings and refine the analysis

Researchers review the sample scope, classification criteria and conclusions.

### Content recommendation · Semantics & audience relationships

PURPOSE & MY CONTRIBUTION

I lead development and upgrades, building semantic queries, audience relationships and ranking based on business criteria. The engine connects topics, use cases and audience concerns to help teams find relevant content and assess topic choices.

Brand · × · Use case · × · Audience concerns

Related content & audiences → Business criteria → Team review

### Related development · LLM applications and content generation

PURPOSE & MY CONTRIBUTION

My LLM application work includes model selection and adaptation, semantic processing, context design and quality checks. In content-generation development, product references and business rules guide generation, while editors review the output.

Product references · × · Business rules · × · Review criteria

Prepare context → Generate → Human review and revision

[Source](https://zhangxin.chat/#work)

BUSINESS APPLICATIONS

## Development and iteration in business settings

Translating client requirements into classification criteria, system rules and workflows. The following projects show my contribution.

A LARGE PHARMACEUTICAL GROUP · Consumer research

### Turning research requirements into system rules

Consumer research on pain-related and other topics, examining needs, usage contexts and product feedback to inform product positioning and opportunity assessment.

Consumer voices · → · Shared criteria · → · Research findings

DELIVERABLES

Multiple research reports and delivery materials during 2025.

MY CONTRIBUTION

Contributed to classification criteria, output requirements and client workflow improvements, translating research requirements into system rules used to prepare reports and analysis.

BIOSTIME · Research iteration

### Adapting the research system to business feedback

Revised the research methods, system and analysis in response to business team feedback on data selection and findings, for continued use and discussion.

DELIVERABLES

Delivered the revised analysis to the business team.

MY CONTRIBUTION

Responsible for data selection, research system updates and analysis improvements, with guidance to the business team on data criteria and use of the findings.

[Source](https://zhangxin.chat/#questions)

POSSIBLE COLLABORATIONS

## Building systems together around business needs

Start with a concrete task, then agree on the technical approach, scope and responsibilities.

### Automated research systems

For recurring research tasks, define the data scope, classification criteria and review requirements, then develop filtering, analysis and reporting workflows. Test the core steps on a real question and iterate with researchers.

Possible outputs · Data and classification rules · Research task orchestration · Source-linked outputs and review

### Recommendation & creation engines

Turn brand, use-case and audience needs into semantic queries and ranking rules. Where generation is needed, include product references, context and review requirements, then develop and test the workflow with editors.

Possible outputs · Semantic queries and ranking · Generation context and product-reference constraints · Human review

### Agents & enterprise applications

Start with a role or task. Design data and tool integration, model calls and orchestration, including human checkpoints and exception handling. We can develop an enterprise agent platform or application together, validating a core workflow before extending the scope.

Possible outputs · System architecture and tool integration · An agent workflow to test · Iteration and handover

[Source](https://zhangxin.chat/#about)

ABOUT ZHANG XIN

## Engineering, entrepreneurship and team leadership

My experience spans more than a decade in engineering and team management, including founding a company and its subsequent acquisition. I remain involved in development, with a focus on how systems fit the team's day-to-day work.

- RECENT WORK LLM applications & AI agents Systems development for research, content and business workflows, including model selection and adaptation, context design, task orchestration and quality checks.

- AFTER ACQUISITION Fanjiedan · CTO Became CTO after the company I founded was acquired by Fanjiedan, working across technical direction, product development and team management.

- FROM 2018 Jile Lab · Founder Founded Jile Lab to develop social-media NLP and data products, including software for brand and market research.

- 2013–2018 Tencent → Meituan Product development, engineering standards and team building, across Tencent Fengling, Meituan transport and smart restaurant products.

[Source](https://zhangxin.chat/#approach)

WORKING TOGETHER

## From a business problem to system development

Start with a small version of a core workflow. Test and improve it with your team, then agree on the next stage.

- 01 Understand the requirements Talk with the people who will use the system. Understand the business goals, available data and current workflow, then choose one task to address first.

- 02 Build a small working version Agree on scope, responsibilities, investment and acceptance criteria. Build one core workflow and test it on real tasks.

- 03 Iterate with feedback Improve the system with user feedback. Agree on the next scope, release cadence and handover before deciding whether to expand.

[Source](https://zhangxin.chat/#contact)

GET IN TOUCH

## The next step. Let’s talk business.

Please contact the friend who shared this page with you. A brief introduction to your business, the issue you want to address and your current approach will help us arrange a useful discussion.

Copy it, add your context and send it to your introducer.

- Your company and business

- The problem you want to address

- Your current approach and resources