ValueGuard
Evidence-based after-sales decisions, with human oversight
My roleTeam leader · Product planning and AI-assisted M2 engine development
Project timeline2026.08 — 2026.09
Evidence of delivery15 · Fixed test cases in a PoC using a real model
VALUEGUARD / 01◇
01Customer signals
02Perception / Attribution / Strategy
03Human confirmation
The problem
Customer value, sentiment, and after-sales information were scattered. A complex problem needed to become an actionable, explainable decision workflow with human oversight.
What I did
- Led product definition and the PRD, structured the experience around a case queue, evidence details, and human confirmation, and broke the work into 4 modules and 37 execution tasks.
- Handled AI-assisted development of the M2 engine, organizing a three-stage agent workflow around perception, attribution, and strategy; constrained outputs through structure validation, evidence citations, and an action allowlist.
- Helped the team integrate React/TypeScript, FastAPI, and SQLite, completing the workflow from case import and analysis to human confirmation, persistence, and reading the stored results back.
Deliverables & results
The team delivered a working Web MVP and 3 Sprint Reports.
The project materials report that all 15 fixed cases in a PoC using a real model succeeded, passing 45 stage-level checks.
What has been validated
A team practice project using processed public data and Mock cases. The PoC demonstrates that the workflow runs for the fixed cases; it does not establish customer retention benefits in a real e-commerce business or production stability.