type
Post
status
Published
date
Nov 18, 2026
slug
summary
tags
category
行业动态
icon
password
48 Hours Inside Xiaohongshu's AI Campus Drive: What Top Tech Talent Really Discussed
Recently, an unusual trend emerged in the Chinese tech recruitment sector: the campus hiring process shifted from formal presentations to immersive, face-to-face technical exchanges. On November 14 and 15, Xiaohongshu held its "REDpath Campus Tech Salon" in its Beijing office. Over 48 hours, nearly 200 candidates from cities like Shanghai, Nanjing, and Hong Kong—along with one international student arriving from Singapore—gathered not just to observe, but to engage in direct dialogue with engineering leaders. The format prioritized an AMA (Ask Me Anything) session over scripted pitches, creating a space where questions about technical direction, team culture, and career growth were answered openly.
AI Is No Longer Optional: It’s the Core of New-Hire Roles
A defining theme of the two days was the inescapable role of Artificial Intelligence. Students were candid about their anxieties: if AI can write code and generate content, what remains for a junior engineer? In the Transaction and Basic Data Warehouse group, one intern expressed fear that by building AI efficiency tools, they might be "training their own obsolescence."
Yet the company’s response was decisive. Xiaohongshu has restructured its job titles; a role previously labeled "Data Development" is now "AI Application and Development." The company introduced a new "Product Engineer" track, where engineers use AI to transcend traditional front-end or back-end silos, participating in the full product cycle from requirement analysis to delivery. To support this, the company issues monthly AI Tokens to all employees, allowing unlimited top-ups. The integration goes deeper than tool usage: even non-technical roles now face written tests requiring candidates to build simple AI demos, signaling that "vibe coding" and AI literacy are baseline expectations.
From "Hammer in Hand" to Intelligent Agent Builders
With routine coding tasks increasingly handled by AI, the focus for human engineers has shifted to building Agents.
The recommendation algorithm team has deployed an interactive recommendation agent. Users can comment directly on posts, saying things like "recommend more like this," and the system adapts in real-time. The Data team is building a Data Agent, where employees develop specialized "skills" and knowledge bases for their domains, then benchmark accuracy to integrate the best tools.
However, technical leaders warned against a blind rush toward large models. Ji Hao, head of video recommendation, emphasized that solving a problem with a massive model consumes significant compute; if a simple strategy yields the same business result, the complexity isn't justified. The key skill, he argued, is knowing *which* tool to apply. This judgment is a core criterion in interviews, where fresh graduates' unbounded creative thinking is valued over rigid adherence to industry frameworks.
New Hires Ship "Works" in Year One
A major draw for candidates is the opportunity to work on high-impact projects immediately. Xiaqi, a Media Lab algorithm engineer fresh out of Tsinghua University, led the image enhancement strategy for the 2026 World Cup. The challenge: football broadcasts feature long shots where players occupy only a few pixels, causing standard algorithms to blur faces and limbs. Her team re-trained models using thousands of match frames to preserve detail, a feat rarely accessible to a first-year employee.
Similarly, Haolin, a 2024 graduate from Peking University, spearheaded the exploration of 3D Spatial Photos. The project involved solving data incompatibility between phone models to allow spatial photos to be shared on the platform. Leadership supported this research despite the risk of failure, restructuring his workload to allow for deep technical exploration. The result: a young engineer shipped a novel feature, proving that the company protects and resources experimental thinking.
Community Values Override Pure Metrics
Perhaps the most striking insight comes from the tension between algorithmic optimization and community health. With 400 million monthly active users, the platform faces a constant challenge: a recommendation strategy that increases watch time might also amplify content that harms the community ecosystem. Ji Hao revealed that Xiaohongshu will sometimes *reduce* the exposure of content, even if it boosts short-term metrics, to align with community values.
This stance is embodied in features like the voice comment function. When users with disabilities began using it to express themselves, the impact went beyond data points. It demonstrated that technology, when guided by human judgment, creates tangible social value.
As one senior engineer, who has been with the company for 5.5 years, put it: the environment allows for "wild growth." The technical problems—whether it's making search understand true user intent or deciding what content to reward in an AI-generated world—remain unsolved. For top talent, the promise isn't just a salary, but a front-row seat to defining the future of social technology.
Source: QbitAI, https://www.qbitai.com/2026/11/510737.html
- 作者:数字Baby
- 链接:myaigc.top/article/41cb7c2d-0e98-81ee-a26e-cc598db90be9
- 声明:本文采用 CC BY-NC-SA 4.0 许可协议,转载请注明出处。



