POI-TECH logo

Full-time

Product Manager, Team Lead

Guangzhou POI-TECH Intelligent Information Technology Co., Ltd.

Building big data products from scratch taught me that the best features are born out of real customer friction. Leading a product team of 4–5 and working with up to 40 engineers across 100+ sprints to scale our SaaS and AI portfolio taught me how to bridge business vision with deep engineering execution.

Product Scope

Product Ecosystem

Owned product strategy for POI-TECH's industrial data portfolio: data integration, metrics, visualisation, AI applications, AIGC services and the MOM platform.

POI-TECH product ecosystem diagram

Product Contributions

Linear RegressionMMOEXGBoost

Industrial AI Operations Research Application

Industrial AI Operations Research Application illustration

Product quality in continuous production depended on operators' experience, which was hard to pass on or scale.

Lead business research and product planning for real-time quality prediction and process optimisation.

Scenario MiningFound high-value production scenarios and turned operators' pain points into measurable model targets.
Model DeliveryWorked with AI engineers on Linear Regression, XGBoost and MMOE models, from training to tuning.
Feedback LoopFed live production data back through OpenAPI to keep refining predictions and recommendations.

Deployed on 23 production lines across 5 enterprises, saving US$3M a year.

  • 90%+ real-time prediction accuracy
  • US$1M+ energy savings in one optimisation scenario
  • Outstanding case at WAIC (World Artificial Intelligence Conference)
RAGSupervised Fine-Tuning

AIGC Application Service

Customers wanted AI answers grounded in their own operating data and paper-industry knowledge, which general LLMs did not have.

Lead the product team to plan and ship AIGC services for operational Q&A and industrial knowledge retrieval.

Market ResearchCompared 6+ mainstream LLMs and defined the product scope end to end.
Knowledge BaseBuilt a RAG pipeline over proprietary documents, IoT data and metrics, with vector retrieval and prompt orchestration.
LaunchCoordinated supervised fine-tuning and shipped AI analysis for mobile and web.

Launched the MVP in 3 months; 10+ enterprises expressed intent to sign.

  • Answers calibrated by domain experts and cross-checked by a second LLM
  • 20+ enterprise expressions of interest
SQLAPIMySQL

Data Integration Tool

Data Integration Tool illustration

Every factory system (IoT, MES, ERP, WMS and more) needed its own integration, so connecting data was slow and costly.

Design a configurable data service covering metric management, data integration and OpenAPI access.

Metric LibraryDefined basic, derived and interface metrics as ready-made inputs for BI and analysis.
Data IntegrationBuilt configurable collection, query, export and compute-task modules for IoT, MOM+, QMS, ERP, MES and WMS sources.
OpenAPIStandardised data access through unified APIs with managed documentation and parameters.

500 API integrations for 20 enterprises in the first year; delivery time per interface cut from 16 hours to 4 or less.

  • 20+ internal-system OpenAPIs launched
  • 90+ metric calculation scenarios covered
  • Most data volume from IoT and sensor collection
HTMLJavaScriptD3.js

Digital Factory

No-code Data Visualisation Product

Digital Factory illustration

Every customer dashboard needed developers, and a full Digital Factory took about 3 months to deliver.

Design a no-code dashboard product that teams assemble from reusable components and templates.

ComponentsDesigned 20+ chart, card, table and widget components for industrial monitoring.
Live DataDefined live updates, linked components, drill-down, filters and data-source management.
TemplatesAdded 24 standard templates, permissions and multi-device preview to speed up delivery.

Turned custom dashboards into self-service: delivering a full Digital Factory dropped from 3 months to 1.

  • 3,000+ dashboards built on the platform
  • 20+ components and 24 standard templates
HTMLJavaScriptD3.js

Reporting Platform

No-code Report Design & Export Tool

Factories still built daily and monthly reports by hand, offline, which was slow and repetitive.

Move report creation online so non-technical users can build and export reports themselves.

Report TypesDesigned templates for classic, Excel-style and business reports.
Report BuilderDefined data selection, condition editing, preview, sharing and export for non-technical users.
Access ControlAdded permissions so each department could manage its own reports.

Made reporting self-serve and removed repetitive monthly report work.

  • 80%+ of report scenarios covered
  • Used by 500+ enterprises
IoTAPSMESEAMWMSBig Data

Process Cloud MOM+ 3.0-4.0

Factory data sat in separate IoT, planning, execution, maintenance and warehouse systems.

Help build one industrial MOM platform with 8+ modules and enterprise-scale data integration.

PlatformWorked with the product team on 8+ modules, including IoT, APS, MES, EAM and WMS.
Big DataWorked with big-data engineers to clean and integrate 100 TB+ of data a year for real-time queries.

When I left, the platform served 500+ enterprises and 20K+ users, with 2,000+ daily active users.

  • 100 TB+ of data integrated a year
  • A shared data foundation for the AI and visualisation products