This guide draws on 9 first-hand interview accounts from Nowcoder. Candidates describe about 1 round on average, most often technical interviews and behavioral interviews.
These 9 writeups cover data, software engineering roles. All of them were posted under a forum handle, so we could not verify the authors’ identities.
New grad. 1 round. Shared by JeffreyLove on nowcoder, 2026-09-14
1First roundTechnicalInternship and agent design deep dive followed by a hand-coded date difference problem.Covered: Agent product design, Agent memory, Vector vs database retrieval, Deployment and QPS, Date difference without libraries
Agent engineer (Qwen)
Outcome not shared
New grad. 1 round. Shared by Decadeqqq on nowcoder, 2026-09-22
1Second interviewTechnicalExtensive questioning on agent evaluation and deployment work, then LLM fundamentals and fit.Covered: Previous internship and non-return, Agent write-operation risk control, Skill evaluation system and tracing, LLM-judged quality scoring, Agent runtime deployment and isolation, Fit for training and inference infrastructure, Transformer and attention internals, Career plans
AI Application Developer
Outcome not shared
New grad. 1 round. Shared by 在抱佛脚的芝士很完美 on nowcoder, 2026-09-15
1First roundTechnicalMySQL, browser and Redis fundamentals followed by an AI coding exercise and a discussion of the approach.Covered: Project problems, Index optimization and EXPLAIN, Leftmost prefix, B+ tree IO, URL to page render, Redis persistence, AI coding task
Role not stated
Outcome not shared
New grad. 3 rounds. Shared by 可乐roue on nowcoder, 2026-09-16
1First roundTechnicalTechnical interview.
2Second roundTechnicalTechnical interview with a second-level manager, followed by a long wait.
3HR roundBehavioralLight conversation on internship growth and plans. Results to be compared with return-offer candidates.Covered: Growth during internships, Challenges, Career plans
Role not stated
Outcome not shared
New grad. 1 round. Shared by Decadeqqq on nowcoder, 2026-09-16
1First roundTechnical75-minute round with a rate-limiter coding task, metrics pipeline design, messaging reliability and an in-depth agent architecture discussion.Covered: Thread-safe sliding window rate limiter, Training cluster metrics design, Distributed aggregation, Delay queues, Kafka message loss, Agent loop and replanning, Tool calling internals, Context assembly and overflow, Model switching mid-session, Reconnect after refresh, Agent evaluation and datasets
AI Infrastructure Engineer
Outcome not shared
unknown. 1 round. Shared by 算法打工人小旺 on nowcoder, 2026-09-17
1First roundTechnicalDeep questions on RL training infrastructure, distributed parallelism and LLM inference serving.Covered: MoE routing replay between rollout and training, Routing data flow and communication cost, NCCL, NVLink, PCIe and RDMA, DP, TP, PP and EP parallelism, Prefill/decode disaggregation, KV cache reuse and transfer, PagedAttention and prefix caching
Agent Developer (Qwen)
Outcome not shared
unknown. 1 round. Shared by 算法打工人阿旺 on nowcoder, 2026-10-04
1Second roundTechnicalCoding a quicksort, a project walkthrough, then a series of agent and LLM tool-use design questions.Covered: Quicksort implementation, Hardest project deep dive, LangGraph components, Tool-calling failure modes, Function calling internals, Query rewriting, Multi-turn agent weaknesses, Agent skills design, Workflow vs agentic architectures, Prompt assembly order
LLM Algorithm Engineer (Qwen consumer app)
Not selected
unknown. 1 round. Shared by 包子馒头热乎饼 on nowcoder, 1970-01-01
1First roundTechnicalA round on reasoning data, reward modeling and data sampling with a probability puzzle and a scenario question. The candidate was rejected.Covered: Internship scenario and work, Long vs short chain-of-thought data, Image-dependent questions, Fixing poor training results, Building image-text data, Process reward model work and training, Agent experience, Coin-toss expected value, Improving bad cases from a large data pool via sampling
AI Infrastructure Engineer
Outcome not shared
unknown. 1 round. Shared by 算法打工人阿旺 on nowcoder, 2026-09-18
1Second roundTechnicalDeep inference-systems round on distributed debugging, kernel determinism and serving architecture.Covered: Debugging TP with CUDA Graph hangs, Minimal reproduction strategy, Deterministic GEMM and split-K, Persistent kernel shape selection, Decode CPU/GPU sync elimination, Ring vs tree AllReduce, Prefill/decode disaggregation and KV cache transfer
What is the interview process like at Alibaba Group Holding?
This guide draws on 9 first-hand interview accounts from Nowcoder. Candidates describe about 1 round on average, most often technical interviews and behavioral interviews.
What topics does Alibaba Group Holding test in interviews?
Commonly reported topics include Career plans, Agent product design, Agent memory, Vector vs database retrieval, Deployment and QPS, Date difference without libraries.
These guides summarize public, first-hand interview experiences shared by candidates. They describe the shape of each loop and the topics that came up, not a leaked question bank, and every experience links back to its original source. Processes change often. Treat this as directional prep, not a script.
TheJobsMap