AI Training Plans: Evidence, Constraints, and Professional Review
Design a reviewable training-plan workflow from longitudinal records, fitness tests, historical plans, and an exercise library.
Peng
I’m Peng, a Java backend developer and the author of TechShare. My project experience includes online classroom systems and student fitness data platforms. I write about Java, distributed systems, databases, and AI engineering, focusing on implementation decisions, troubleshooting, and practical trade-offs. Articles distinguish firsthand project experience from illustrative examples and explanations based on technical documentation.
Beyond Uptime: Redis Pressure and Classroom Reliability
Measure online classroom reliability through Redis retention, ClickHouse event definitions, entry latency, and after-class archival backlog—not uptime alone.
Online Classroom: Media Failover and Courseware Fallback
A practical online classroom architecture for media failover and courseware fallback, with clear recovery boundaries for Agora, ZEGO, classroom state, and post-lesson replay.
Classroom Messaging: Ordering, Deduplication, Reconnection
Trace how regional classroom messaging keeps pages, annotations, and teacher actions consistent across delay, duplication, route failover, and student reconnection.
QLoRA: Classify Support Tickets with a Small Experiment
A small QLoRA experiment for support-ticket classification, covering configuration, observed results, and the limits of loss-based evaluation.
Transformers: Self-Attention to Decoder-Only Models
Explain self-attention, decoder-only transformer blocks, and the checks needed to connect model architecture with real inference behavior.
Text Embeddings for RAG: From Token Lookup to Retrieval
Explain how token lookup becomes a retrieval vector, then test ranking quality, encoding choices, and retrieval-model evaluation.
Reliable RAG: Document Quality, Retrieval, and Evidence
Build a support assistant around document quality, access controls, policy versions, hybrid retrieval, and citation checks.
Spring AI Alibaba Support Assistant: Tools, Memory, and RAG
Build a support assistant with Spring AI Alibaba and Qwen while keeping identity, authorization, memory, tools, and policy retrieval under application control.
SUM(DISTINCT) vs JOIN Duplicates: Keep Payment Identity
Learn why SUM(DISTINCT amount) can discard valid payments, how SQL deduplicates rows, and when to use payment identity, EXISTS, or explicit allocation.