01 / The Philosophy

Focused on AI Automation System Delivery
building enterprise digital solutions that actually run in production, not half-finished demos.

I'm Heron, an FDE (Full-stack Development Engineer) focused on AI automation system delivery and enterprise digital solutions. There's no shortage of people who can talk about AI concepts — what's scarce is someone who can independently deliver top-level intelligent workflows as a complete, closed-loop system that actually runs inside your business, and who takes responsibility for the outcome. That's the core of what I do.

Traditional AI rollouts often get stuck in an awkward spot: the technical approach makes sense on paper, but nobody owns wiring it into the real business workflow. My core value lies in breaking down system silos — embedding AI automation capability directly into your existing fingerprint-browser environments, internal ERP systems, and Feishu office workflows, so the automation actually runs and compounds into a reusable digital asset for your business, not a one-off proof of concept.

The Disadvantage of Split-Stack Development

🤝 Interaction Disconnected From the Backend

Backend-only agents lead to clunky front-end interaction, unable to support the complex dynamic flows and real-time data visualization that multi-agent coordination requires.

🚧 Poor Business Fit

Generic SaaS on the market can't adapt to your unique business processes (SOPs), leaving staff to manually copy-paste between disconnected systems.

🔒 Critical Data Privacy Exposure

Connecting core business assets — finances, customers, proprietary tech — to the public cloud is like going unprotected. Secure, controllable, full-stack private deployment is the only real way out.

💸 Unsustainable Coordination Costs

Outsourced teams with separate frontend, backend, and AI engineers rarely agree, communication chains get long, and projects slip or collapse. A full-stack solo operator is faster and more stable.

FDE Delivery Standard: High-Fidelity Frontend + Industrial-Grade Automation Loop
From the top-level responsive UI (React/Next.js/Tailwind), through the mid-tier high-performance business gateway (Flask/Node.js), down to deeply integrated high-concurrency multi-agent clusters and webhooks. You keep 100% data sovereignty — it's your own automation hub, entirely.

02 / FDE Full-Stack & AI Services

Custom E-commerce System Overhaul

Deep integration with mainstream e-commerce backends and internal data interfaces like Jijia ERP, delivering a tailored closed-loop control panel with automated risk filtering and analytics.

Private Deployment of OpenClaw Architecture

Full deployment of the OpenClaw architecture on your physical servers or cloud environment. Supports fine-grained agent prompt engineering against local knowledge bases or dedicated lines, keeping core business data 100% isolated.

Deep Integration Into Your Collaboration Stack

Break down ERP data silos. High-performance Python/Flask interface services embed business dashboards seamlessly into Feishu, WeChat, and other office tools, with dynamic interactive card pushes and approval workflows.

Headless Browser & Anti-Detection Collection

Built for high-concurrency cross-border risk-control scenarios, integrating anti-association browsers like AdsPower/Ziniao with Playwright for efficient competitor and price-defense monitoring down to ±5% granularity.

03 / The Tactical Toolset

*Spanning frontend, backend, and automation technologies to keep systems running at industrial-grade availability.

OpenClaw / Coze Workflow Playwright + AdsPower/Ziniao Feishu Interactive Cards Python Fullstack / Flask Next.js / React Jijia ERP API Auth Tailwind CSS / TypeScript RAG / Vector Database

04 / Featured Full-Stack Workflows

Cross-Border E-commerce Intelligent Dispatch | Amazon Ads & Competitor Defense

Dual-Line Defense Hub for Ads & Competitor Pricing

Covers 10 core SKUs · Fully automated daily reports across 8 advertising metrics · Real-time alerts for competitor price swings of ±5%

Core pain point: Operators had to manually cross-check ads dashboards, inventory sheets, and competitor pages every day. When ad spend is disconnected from inventory, campaigns can oversell or keep spending into stockouts; missing the response window in a price war means direct profit loss.
Full-stack solution: The daily ads script pulls both "operations data" and "inventory tracking" Feishu Bitable tables, automatically matching stock by SKU. It prioritizes FBA inventory, switches to Winit warehouse stock when FBA is zero, and aggregates inventory across child SKUs with the same prefix, such as color variants, so management no longer has to total them by hand. On the competitor side, a Ziniao fingerprint browser plus Playwright takes over a real browser environment to collect Amazon storefront prices, compares them against the previous day's history, highlights swings of ≥±5%, and pushes Feishu cards automatically.

Enterprise Inventory & Logistics Risk Control | Inventory Health & Logistics Alerts

End-to-End Inventory and Logistics Risk Control System

Automatically locks the TOP 5 high-risk delayed orders · Tiered aging alerts using 3-month and 6-month thresholds

Core pain point: Long-cycle inventory pressure and logistics delays are often discovered only at month-end. Warehouse, logistics, and operations data sit in separate spreadsheets, so abnormal orders are easily buried under normal records.
Full-stack solution: The logistics delay script calculates delay days from the planned warehouse arrival date and the current date. After excluding orders that have already arrived, it pushes the TOP 5 delayed orders in descending order and mentions the responsible owner, so management sees the most urgent cases first instead of scanning dozens of rows. The inventory-aging side monitors both 3-month and 6-month overdue tables, combining them with weekly product value reports to show which goods have been sitting for how long and how much capital they represent. Everything is pushed automatically on daily or weekly schedules, with no manual lookup or consolidation required.

Engineering Resilience Case | When the Official API Breaks

After the Official API Failed: RPA Seamlessly Took Over Negative Review Monitoring

Switched to an RPA fallback on the same day the ERP gateway authentication failed, with zero monitoring downtime

Core pain point: Negative review monitoring originally depended on the official Jijia ERP API, but its gateway authentication failed and stayed unavailable for an extended period. This is a common supply-break risk in automation systems that depend on third-party interfaces, and many solutions simply stop at this point.
Full-stack solution: Instead of waiting for the official fix, the data collection layer was switched to Playwright RPA, using a simulated real login session to collect negative review data from the ERP frontend. The upper-layer Feishu card generation and push logic was reused unchanged, so the business side barely felt the switch. This is the practical difference between a full-stack engineer and someone who can only call APIs: an interface can stop, but business monitoring cannot stop with it.

Cross-Department Operations Data Hub | Multi-Report Automation Matrix

Automated Operations Reporting Matrix: Ads / Sales / Logistics / Aging / Competitors / Reviews

Unified hub for 7 categories of daily reports, weekly reports, and alerts · Supports --dry-run previews for acceptance without touching live group chats

Core pain point: Every day, the team had to manually reconcile ad spend, inventory, logistics timeliness, competitor prices, and other data across Feishu Bitable, the ERP backend, and Amazon Seller Central. Missing any one category could lead to stockouts or tied-up capital, while manual daily reports were slow and easy to lag behind the business.
Full-stack solution: A complete scheduled-script matrix was delivered. It pulls structured data from Feishu Bitable, fills in no-official-API collection scenarios with Playwright plus fingerprint browsers, then consolidates everything into Feishu Interactive Cards pushed to the relevant owner groups: daily ads reports, daily sales reports, air freight logistics summaries, logistics delay alerts, overdue inventory aging statistics, weekly product value reports, competitor price fluctuation monitoring, and ERP negative review monitoring. All scripts support --dry-run mode, making it possible to preview outputs without touching live group chats, which keeps iteration and acceptance clean.
Engineering details: The collection layer and push layer are decoupled, with scripts split into scrape_*.py and report_*.py, so a single data-source failure does not block the remaining reports from being delivered on time. One script also uses Playwright to operate the Feishu Bitable connector configuration page, because Feishu does not expose an official API for that part. It automatically changes the "query time" parameter to the previous day and saves it, turning another formerly manual click path into an unattended workflow.

05 / Workflow at a Glance

Behind these cases is the same reusable closed-loop methodology. Whether the business scenario is ad bidding or inventory liquidation, the delivered system follows the same four-stage structure:

FDE full-stack delivery workflow diagram Data collection flows into agent decision-making and then business execution. Exceptions or high-risk cases return to human confirmation to form a closed loop. 01 · Data Collection Playwright + Fingerprint Browser / ERP API 02 · Agent Decision OpenClaw Multi-Agent SOP-Based Rules 03 · Business Execution Pricing / Pause / Clearance Write Back to Systems 04 · Human Loop Feishu Interactive Cards Owner One-Click Confirm Dashed line: exceptions and high-risk decisions return to human confirmation instead of auto-executing.

The collection layer keeps data faithful, the decision layer handles explainable rule and model judgment, the execution layer writes directly back into existing enterprise systems instead of creating another silo, and the loop layer leaves final discretion with the business owner. That is the essential difference between full-stack delivery and simply connecting a single API.

06 / Frequently Asked Questions

Do you support private/on-premise deployment?

Yes. Systems can be deployed to your company's own servers, a private cloud, or a local network environment, so business data never passes through a third-party platform.

Do we need to switch our existing ERP?

No. Integration is prioritized with your existing ERP, Feishu, WeCom, and other systems already in use, keeping business changes to a minimum.

How long does delivery take?

Typically 2–8 weeks depending on project complexity. A phased delivery plan is provided once requirements are confirmed.

Can you sign an NDA?

Yes. For projects involving sensitive business data or workflows, a non-disclosure agreement can be signed.

Is ongoing support provided?

Deployment support, bug fixes, and maintenance are provided as agreed, with long-term technical support available depending on the project.

How is pricing determined?

Pricing is based on requirement complexity, system scale, and delivery scope. A free 15-minute consultation call is available first.

07 / Client Project Feedback

Anonymous feedback from real project deliveries. Company names are withheld per NDA agreements.

Shenzhen · Consumer Electronics · GMV ~¥30M
Ads & Inventory Linked Report
Ops team 20+ · 12 Amazon stores
Amazon Playwright Feishu Jijia ERP 3 Weeks

45min→3min Report Gen Time
12 Stores Covered
~2h/day Labor Saved
The daily report auto-links inventory — uses Winit when FBA is out. Details like this show they actually understand how our business works, not just pile up numbers.
Shenzhen · Outdoor Gear · GMV ~¥50M
Competitor Price Monitoring & Alerts
Ops team 30+ · 50+ core SKUs
Amazon AdsPower Playwright Feishu 2 Weeks

±5% Alert Threshold
<5min Response Latency
↓95% Manual Monitoring Cut
Price changes over 5% auto-trigger alerts — no more manual price checking every day. "Alert on exception only" removes so much noise.
Guangzhou · Home Furnishing · GMV ~¥80M
Inventory & Logistics Risk Control
Supply chain team 50+ · Global multi-warehouse
Feishu Python OpenClaw Winit 4 Weeks

TOP 5 Auto-Lock High-Risk Orders
3/6 Mo Dual-Threshold Aging Alerts
100% In-Stock SKU Coverage
Weekly-level data like inventory aging and product value is also covered. It doesn't just focus on short-term sales — the long-term health of inventory is included too.
Shenzhen · 3C Electronics · GMV ~¥200M
ERP Automation & RPA Emergency Fallback
Ops team 100+ · Multi-brand management
Playwright Jijia ERP Feishu RPA 2 Weeks

<24h API Down → RPA Switch
0 Days Monitoring Downtime
99%+ Data Capture Rate
Didn't wait for the official API fix — switched the data layer to RPA within a day. Upstream notification logic stayed untouched, and the business team barely noticed.
Cross-Border Group · GMV ~¥500M
OpenClaw Private Deployment & Ops Hub
Full team 200+ · Multi-system integration
OpenClaw Private AI Feishu RAG 6 Weeks

100% Data Physically Isolated
7 Types Auto Reports Pushed
↓80% Manual Work Reduced
Open Feishu in the morning and see everything — no more logging into four or five platforms. Zero learning curve for the team.