AI Alumni App, Beijing Powersoft Technology
Product Manager
Rebuilding how Chinese universities keep and use alumni records at 100k+ scale.
Abstract
Chinese universities rely on fragmented alumni records scattered across WeChat groups, mini-apps, and inconsistent spreadsheets. Without a single source of truth and only relying on self reported majors and industries, the data is messy and impossible to organize at scale for fundraising, career services, or community building.
The challenge was to design a system that served both the customer, universities and the end user, alumni, and to balance the needs of these two key users who are currently unsupported by existing tools.
I ran customer interviews, benchmarked competitors, and walked through how the alumni offices actually worked day to day. Because this was B2B I spent as much time on what the staff had to maintain, the tagging and the cleanup and whether they believed the system, as on what alumni would see.
- ·Interviewed ~7 alumni staff and alumni, who revealed that the main challenge was not lack of alumni enagemenet, but more about the data quality and the ability to search and find people.
- ·Conducted competitive analysis:
- –Local CRMs: better workflows but weak AI
- –Western tools: better UX but no WeChat support or local visibility
- –Most schools still relied on manual questionnaires
- ·Validated natural language search over complicated filtes after interviews.
Going to such a prestigious school but not being able to utilize the network beyond those in your year was disappointing.
Approach & Strategy
To balance the needs of university customers and alumni end users, I identified three key gaps to build toward:
- ·When alumni appear in the news, match those mentions to the right person in the database so classmates can follow what graduates are doing after school.
- ·Clean up messy, self-reported major names using AI normalization so the same department is not scattered across dozens of variants in search and reporting.
- ·An AI chatbot that supports natural language search instead of complicated filtering.
Key Tradeoffs
- 1.Rule-based matching vs. AI normalization for matching alumni in scraped news to the database. Rule-based matching would have shipped faster, but we chose accuracy over speed because trust was everything for our customers. Schools had to be able to trust the data completely and even one bad match could undermine confidence in our product.
- 2.Natural language chatbot vs. structured filters. Filters are predictable, but they create friction when someone doesn't know what they are looking for or how the data is organized. We put chatbot-guided search in front, with structured alumni data still underneath.
An Alumni AI Mini-App on WeChat and web, where alumni got AI-curated news about classmates, browsed industry-tagged profiles, and searched in plain language. On the backend we replaced the brittle rule-based workflows with AI matching and normalization.
The chatbot used GenAI with RAG grounded in the alumni profiles and news data, which mitigated hallucination.
The entity matching pipeline had to scale with 200k records per school. I built and tested a normalization pipeline that ran text cleaning, synonym mapping, blocking strategies, similarity scoring, and threshold clustering.
Impact
- ·Raised entity-matching accuracy from ~80% to 99%, making alumni news and profile data trustworthy.
- ·Deployed across 5+ universities, confirming the need was repeatable beyond a single school.
- ·Replaced ad-hoc spreadsheets with an organized, searchable system for schools and gave alumni faster ways to find people and track career updates.