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.
🤝 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.
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.
04 / Featured Full-Stack Workflows
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.
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.
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.
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:
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.