Kaison Lau · Forward Deployed Engineer
中文

Kaison Lau Mingxuan Liu

Target: Forward Deployed Engineer (FDE)

Bachelor (dual degree) · Wuhan, China · 3 years · open to travel / on-site

+86-19893173727 · i@41box.com · github.com/kaisonlau8 · lancloudtech.com

Kaison Lau

Profile

I do the FDE job: embed with the people who run the business, turn an ambiguous pain into a shippable wedge, write the production code, and stay until it runs every day. Success is adoption and a workflow change — not a demo — plus patterns that can go back into the product. By day I sit in Dongfeng M-Hero after-sales operations, next to supervisors, stores, and DMS. In parallel I productized what worked as LAN Cloud AI (LeadsHunter in production). Full-stack Python / TypeScript, LLMs in real workflows, comfortable in rooms with operators, IT, and executives.

Education

Lanzhou Jiaotong University 2018.09 - 2023.06

Vehicle Engineering (major) + Accounting (minor) · B.Eng. + B.Mgmt.

CET-4 English. Designated English host for overseas VIP visits at the M-Hero Experience Center (26th World Youth Congress; 2026 HUST International Summer School).

Experience

Dongfeng Motor Group — M-Hero Automotive Technology 2026.01 - Present

Delivery Service · After-sales operations, embedded delivery

How I work

  • Sit in service-ops meetings and turn inspections, complaints, NSS, DMS, and the store network into systems with an owner, a clock, evidence, and a risk.
  • Work with regional supervisors and 120 authorized stores: travel, on-site inspection, staffing — tools are built beside the people who use them.
  • Integrate inside the customer environment: mapped 158 DMS pages (114 with data); Feishu bitable / bots and WeCom SCRM; session keepalive → export → match → alert.
  • Put LLMs into existing work: weekly complaint merge → summary / risk tag / inspection link / disposition; kick off with business, data, IT, and AI — not a slide-only analysis.

In production

  1. Accident-repair overdue alerts live: 7/10/14-day rules, 10:00 store/region push, 17:00 national KPI. VIP maintenance reminders use the same DMS → match → Feishu loop.
  2. SkyEye / Dongfeng Eye: 27-table layer + post-level sentiment/risk agent. Group AI contest, 7-minute briefing on 2026.07.07; Young Role Model and 1st place in the youth skills contest.
  3. 120-store / 29-province capability survey: tire-repair self-run rate 25.8%; 31.7% of stores with 2+ gaps; short/mid/long remediation, not a one-page conclusion.
  4. Complaint governance as P0/P1/P2 (structure & triage → suggestions → monthly root cause), matched to store inspections for a reviewable loop.
Customer-embedded Production delivery Enterprise integration LLM in the workflow
Sichuan LAN Cloud AI Technology Co., Ltd. 2026.07 - Present (concurrent)

Founder · turn field-proven loops into products

  • Method: business rules × bitable / event systems × AI. Prove in real stores, then productize behind contracts.
  • LeadsHunter: Mercury (v1.0.12) is in production (ingest → agent score → multi-tenant delivery). Venus (v2.1, in development) adds a three-pool ledger and a read-only sales app (home / leads / notifications, Getui, 10:00 digest). Android beta is out; not cut over to production yet.
  • Also shipped TACT Phase 0 contracts, Cloud Ledger (Workers / D1 / R2), the site and official mini program, and a fission mall — lancloudtech.com.
Technical founder Pattern capture Production agents
Dongfeng Honda Automobile Co., Ltd. 2024.07 - 2025.12

After-sales · dealer management · tools supervisors actually open

  • 600+ dealers nationwide: throughput, orders, output value, profit, CSI, regional ops, churn — for targets and supervisor action.
  • Built daily output-value tracking, churn heatmaps, and closed-store transfer analysis so regions stopped waiting on manual exports.
  • Recall-lead app and WeCom retention groundwork; Dongfeng Honda 2025 Value Creation Bronze.
Dongfeng Honda Automobile Co., Ltd. 2023.07 - 2024.07

NEV body-shop digital lead · on-the-line delivery

  • Full IT lifecycle on the welding line: sourcing, supplier review, specs, plan, risk, acceptance, floor training.
  • WIMP: execution, VIN trace, work orders, process quality, equipment. 30+ AGVs commissioned; 500+ parts moved flexibly per day; line started on schedule.

Selected work

LeadsHunter · Venus three-pool + sales app 2026

Mercury in production · Venus v2.1 / APP v2.1.4-dev (beta, not production)

Dealers are not short of public conversation. They are short of an action that lands on a salesperson’s phone. Venus files leads through three pools; sales no longer use the web — only a read-only app of what was assigned to them.

  • Three pools: ownership (which org), org-assign (six daily quota buckets), account-dispatch (group-bound accounts). Admins operate on the web; users only read in the app.
  • Separate Expo app (iOS / Android): login, comment/post home counts, lead list and detail, notification center, Getui passthrough, 10:00 digest, jump to the original Douyin / Xiaohongshu post. Android beta APK shipped.
  • Scoring agent on comment / post pipelines (intent, summary, evidence, next step); training-ground skills sync back. Venus runtime retired outbound Feishu in favor of the app and account short links.

PostgreSQL / Node / Vue / Expo RN / Python agent / Getui / TikHub

SkyEye / Dongfeng Eye 2026

Product / full-stack / briefed to the group

OEMs cannot see brand risk forming in public. Land post-level judgment and a disposition surface — defend with a running system, not a concept deck.

  • 27-table data layer; agent for sentiment, risk, brand, topic; risk matrix and brand-health board.
  • Group AI contest entry; Young Role Model; 1st in the youth skills contest.

TikHub / PostgreSQL / Fastify / Vue3 / ECharts / DeepSeek agent / Docker

Accident overdue alerts · VIP maintenance nudges 2026

On-site rules / DMS integration / daily production jobs

The car is not late with the customer yet. Delay is already in the work order. Encode who must see which evidence at which time, and send it to the right role.

  • DMS keepalive and export → tiered rules → Feishu / Excel. Accident jobs at 10:00 / 17:00; VIP reminders de-duplicated by task code.
  • Console yellow pages and port heartbeat so the loop is still up on Monday.

Python / DMS / rules / Feishu bots / launchd

Complaint-governance AI pipeline 2026

Metric design / Bailian summaries / exec kickoff

Complaints live in weekly files; humans cannot close the loop. Lock What-Why-How and 5M1E first, then let the model summarize, tag risk, attach inspections, and suggest disposition.

  • P0/P1/P2 priorities briefed to business, data, IT, and AI. Jan–May lists matched to store inspections and into the quarterly report.

DMS / DashScope Bailian / risk tags / Feishu Bitable

Skills

Field delivery
Problem decomposition, floor / supervisor / exec alignment, travel and on-site work, turning one-off scripts into reusable loops, explaining trade-offs to non-engineers
AI / agents
Production prompts and structured output; comment/post scoring agents; summary / classify / risk / next action; training-ground skills; DashScope / DeepSeek / OpenAI / Anthropic APIs
Full-stack
Python (FastAPI, scrapers, agents), TypeScript / JavaScript (Node, Hono, Fastify), Vue3, React, Expo
Data & integration
PostgreSQL, metric design, DMS / Feishu / WeCom SCRM, REST, auth and session keepalive, idempotent delivery
Ship & watch
Docker, Cloudflare Workers / Access / D1 / R2, Nginx, GitHub Actions, cron / launchd, daily alerts

Awards

Note

I do not start from a platform and hunt for a use case. I stand next to stores and supervisors and encode who must see what, at what time, into a system that runs every day. I can own discovery through on-call, and I will fold repeatable work back into a product (LAN Cloud). English is sufficient for international hosting. I want an FDE seat that embeds on-site, ships production code, and measures adoption and workflow change.