Author

Xiaohongshu Tech REDtech

Official account of the Xiaohongshu tech team, sharing tech innovations and problem insights, advancing together.

139

Articles

Likes

1.2k

Views

Comments

Latest from Xiaohongshu Tech REDtech

100 recent articles max

Xiaohongshu Tech REDtech

Jul 24, 2026 · Artificial Intelligence

UniNote: Unifying Multimodal Representation and Ranking in a Single Embedding Model

UniNote introduces a unified multimodal embedding model that combines representation learning and ranking optimization in a single forward pass, using a two‑stage SFT‑then‑RL training paradigm and Matryoshka Representation Learning to achieve competitive Item‑to‑Item retrieval performance while reducing latency.

Item2ItemMatryoshka Representation LearningMultimodal Retrieval

0 likes · 12 min read

UniNote: Unifying Multimodal Representation and Ranking in a Single Embedding Model

Xiaohongshu Tech REDtech

Jul 23, 2026 · Artificial Intelligence

HELMSMAN: Redefining Large-Scale Vector Retrieval on All‑Flash Servers (OSDI 2026)

HELMSMAN replaces DRAM‑heavy graph indexes with a clustering‑based, SSD‑first ANN system that uses a custom SPDK storage stack, adaptive LLSP pruning, and a GPU‑CPU construction pipeline, achieving 2‑16× throughput, up to 85% of in‑memory performance, and over 90% hardware cost reduction for billion‑scale search workloads.

Approximate Nearest NeighborClusteringGPU

0 likes · 13 min read

HELMSMAN: Redefining Large-Scale Vector Retrieval on All‑Flash Servers (OSDI 2026)

Xiaohongshu Tech REDtech

Jul 22, 2026 · Artificial Intelligence

BigMac: Breaking the Pareto Frontier of Compute‑Memory Trade‑offs in Multimodal LLM Training

BigMac introduces a dependency‑safe nested pipeline that preserves LLM compute efficiency while reducing encoder and generator activation memory to O(1), delivering 1.08‑1.9× speedup and stable memory usage for large‑scale multimodal model training.

Nested Pipelinecompute efficiencydistributed training

0 likes · 15 min read

BigMac: Breaking the Pareto Frontier of Compute‑Memory Trade‑offs in Multimodal LLM Training

Xiaohongshu Tech REDtech

Jul 16, 2026 · Artificial Intelligence

Cut First‑Token Latency by 3.25×: Introducing HYPIC for Position‑Independent Caching in Hybrid‑Attention LLMs

HYPIC combines position‑independent caching with hybrid‑attention LLMs, reducing first‑token latency by 3.25× and sustaining QPS by 1.66× while keeping quality loss under 2 points, through a segment‑cumulative transition operator, a seam‑window fix for full‑attention layers, and segment‑parallel execution.

HYPICKV cacheLLM inference

0 likes · 13 min read

Cut First‑Token Latency by 3.25×: Introducing HYPIC for Position‑Independent Caching in Hybrid‑Attention LLMs

Xiaohongshu Tech REDtech

Jul 14, 2026 · Artificial Intelligence

How Xiaohongshu Built an Enterprise AI Personal Assistant from Zero to Full‑Staff Coverage

Xiaohongshu’s AI team describes a month‑long, three‑person effort that leveraged an AI‑Native project model, isolated Kubernetes clusters, a custom sandbox (NEX), token‑saving Self‑GC, cost‑aware routing, and a three‑layer memory architecture to roll out a secure, low‑cost AI personal assistant used by every employee.

AI AgentEnterprise AISkill-Hub

0 likes · 14 min read

How Xiaohongshu Built an Enterprise AI Personal Assistant from Zero to Full‑Staff Coverage

Xiaohongshu Tech REDtech

Jul 10, 2026 · Artificial Intelligence

PIPO: Pair‑In Pair‑Out Architecture Cuts Input Length in Half and Doubles Token Output for LLMs

The PIPO (Pair‑In, Pair‑Out) architecture compresses two input tokens into one latent vector and expands each hidden state to emit an extra draft token, halving effective input length and doubling per‑step output, achieving up to +7.15 pass@4 improvement and 2.64× TTFT / 2.07× TPOT speedups on benchmarks such as AIME 2025, GPQA‑Diamond, LiveCodeBench v6 and LongBench v2 with Qwen3.5‑4B/9B backbones.

EfficiencyMulti-Token Predictionlatent compression

0 likes · 14 min read

PIPO: Pair‑In Pair‑Out Architecture Cuts Input Length in Half and Doubles Token Output for LLMs

Xiaohongshu Tech REDtech

Jun 24, 2026 · Artificial Intelligence

Scaling Enterprise AI Agents: Self‑GC Context Governance and NEX Sandbox

The AICon 2026 talks detail how Xiaohongshu’s engineering team tackles long‑context agent challenges by introducing Self‑GC’s ContextObject model, side‑channel commit mechanisms, a NEX sandbox for runtime isolation, and an L0‑L2 memory hierarchy, achieving 10‑20% token savings with over 90% impact‑free execution.

AI AgentsContext GovernanceEnterprise AI

0 likes · 7 min read

Scaling Enterprise AI Agents: Self‑GC Context Governance and NEX Sandbox

Xiaohongshu Tech REDtech

Jun 23, 2026 · Artificial Intelligence

How AI Agents Can Slash Server‑Side End‑to‑End Test Costs

The article analyzes why server‑side end‑to‑end testing has become costly due to cross‑domain complexity, long chains, and combinatorial explosion, and demonstrates how an AI‑driven agent with dynamic planning, progressive knowledge‑base loading, and a debug‑first script generation loop can dramatically reduce automation effort from days to minutes.

AI AgentDebug-firstKnowledge Base

0 likes · 17 min read

How AI Agents Can Slash Server‑Side End‑to‑End Test Costs

Xiaohongshu Tech REDtech

Jun 17, 2026 · Artificial Intelligence

RedParrot’s Semantic Cache Accelerates Enterprise NL‑to‑DSL Analytics by 3.6×

RedParrot introduces a query‑semantic‑caching framework that compresses the multi‑stage LLM NL‑to‑DSL workflow into a short‑chain process, achieving an average 3.6× inference speedup and an 8.26% accuracy gain on real‑world business data while also delivering strong generalization on open NL‑to‑DSL benchmarks.

Business AnalyticsLLMNL-to-DSL

0 likes · 19 min read

RedParrot’s Semantic Cache Accelerates Enterprise NL‑to‑DSL Analytics by 3.6×

Xiaohongshu Tech REDtech

May 27, 2026 · Cloud Native

How RedProcess Evolved into DES: Optimizing Xiaohongshu’s Multimedia Task Scheduler

The article details the evolution from the first‑generation RedProcess scheduler to the Distributed Execution Scheduler (DES), explaining how architectural redesigns in storage layering, push‑based dispatch, and systematic disaster‑recovery transformed Xiaohongshu’s video‑cloud task scheduling from merely usable to highly efficient and resilient.

DESKubernetesRedis

0 likes · 15 min read

How RedProcess Evolved into DES: Optimizing Xiaohongshu’s Multimedia Task Scheduler