On July 12, 2025, the DataFun Summit will feature a technical session where China Telecom AI Research Institute engineer Yao Yitong presents a deep dive into enhancing large language model Function Call abilities through systematic data and training optimizations, offering practical insights for AI practitioners.
This article presents a comprehensive overview of the AI‑driven digital‑human live‑streaming solution used by Taobao, detailing six core components—including LLM‑based content generation and interaction, TTS, visual driving, audio‑video engineering, and backend services—while sharing architectural diagrams, cost‑reduction strategies, productization insights, and future directions.
This article presents a comprehensive practice summary of building an intelligent digital‑human system, covering six core modules—LLM content generation, LLM interaction, TTS synthesis, visual driving, audio‑video engineering, and backend services—while detailing data collection, signal processing, ASR annotation, speaker clustering, model optimization (V1‑V4), evaluation metrics, and future research directions.
Model Context Protocol (MCP), an open-source standard from Anthropic, standardizes how large language models interact with external tools and data sources, introducing a client‑server architecture with hosts, clients, and servers, and promises to simplify AI application development compared to traditional function‑calling approaches.
GPULlama3.java, released by Manchester University's Beehive Lab, is the first native Java implementation of Llama 3 that leverages TornadoVM to automatically accelerate inference on GPUs without writing CUDA or native code, supporting NVIDIA, Intel and Apple Silicon back‑ends and modern Java 21 features.
This article introduces a variable‑length chain‑of‑thought distillation technique built on Alibaba Cloud PAI’s EasyDistill toolkit, presents the high‑quality OmniThought‑0528 dataset, details the training of the DistillQwen‑ThoughtY 4B/8B/32B models, and provides code and usage examples for researchers and practitioners.
This article outlines the end‑to‑end architecture and practical solutions behind creating intelligent digital humans for live commerce, covering LLM‑driven content generation, real‑time lip‑sync, image‑driven avatar creation, automated material review, lightweight model training, and a roadmap toward fully automated, high‑performance virtual presenters.
This tutorial walks through the limitations of linear LangChain workflows, introduces LangGraph’s state‑node‑edge architecture, and provides step‑by‑step code examples—including a Hello‑World tool, conditional branching, multi‑turn conversation handling, and graph visualization—so readers can construct robust, persistent LLM agents.
This guide walks you through installing and configuring Dify—a open‑source LLM application platform—on your local machine using Docker, integrating it with Ollama for custom models, and exploring its core features such as chat assistants, agents, workflows, and tool extensions, all illustrated with step‑by‑step screenshots and code snippets.
The article explains the Generator‑Verifier Gap (GVG)—the asymmetry where verifying a solution is far cheaper than generating it—covers its origin, its impact on test‑time scaling for large language models, reinforcement‑learning approaches, and how the concept can shape agent architectures and AI product strategy.
This article walks through the mechanics of greedy search and beam search in large language models, demonstrates both methods with GPT‑2 on the prompt "I have a dream", visualizes the decoding trees, compares their scores, and discusses the trade‑offs between efficiency and output quality.
The article explains what inference engines are, why they are needed beyond raw Python scripts, and outlines best practices such as model quantization, batching, and parallelism, while comparing popular open‑source and commercial options for production AI workloads.
This article details the design, technology choices, functional modules, core implementation, performance optimizations, and future roadmap of Ops‑Nexus, an AI‑driven intelligent operations platform that streamlines alarm analysis, log processing, and host health checks for large‑scale monitoring environments.
This article walks through GraphRAG’s retrieval phase, showing how knowledge‑graph entities, relationships, and community reports are assembled into a query context, comparing local and global modes with traditional RAG, and illustrating the process with a concrete “Age of Big Data” example.
This guide breaks down the ten core building blocks of a production‑ready RAG pipeline—from input handling and vector stores to prompt engineering, LLM inference, observability, and evaluation—showing why each piece matters, common pitfalls, and practical best‑practice recommendations.
This tutorial walks through creating a natural‑language‑to‑SQL chatbot using Spring AI, configuring a MySQL school database with Flyway, defining system prompts for a DeepSeek LLM, implementing service beans and a REST API, and interacting with the bot via curl commands.
The Model Context Protocol (MCP) is an open, standardized adapter that lets large language models seamlessly connect to tools, data sources, and workflows, offering plug‑and‑play intelligence, cross‑platform compatibility, security, and modular extensibility for building real‑world AI applications.
The EXPRESS 2025 workshop at ISSTA in Norway will showcase AI‑driven code auditing, present cutting‑edge research on trustworthy software systems, and invite researchers and practitioners to discuss transparency, reliability, and security challenges in modern software engineering.
The article reflects on Andrej Karpathy’s AI Startup School talk, outlining the evolution from traditional if‑else programming (Software 1.0) through data‑driven models (Software 2.0) to goal‑oriented natural‑language agents (Software 3.0), and examines LLMs as operating‑system‑like infrastructure, prompting, and engineering challenges.
This article details the end‑to‑end AI pipeline for creating intelligent digital humans in live streaming, covering LLM‑driven script generation, multimodal data integration, error‑prone number handling, DPO fine‑tuning, experimental results, and future directions for more human‑like presentations.
This article introduces Eino, an open‑source Golang framework for large‑model AI applications, explains its core capabilities, walks through creating a simple AI assistant with message templates and chat model integration, and demonstrates how to extend the system with tools and a modular architecture for future expansion.
In a detailed interview, Vivo AI engineer Liang Tianan explains how the BlueHeart Small V assistant leverages large language models, multi‑stage recall, ranking, and reward‑model fine‑tuning (SFT/DPO) to generate high‑quality, diverse post‑dialogue recommendation items while balancing latency, cost, and evaluation challenges.
The article explains the Model Context Protocol (MCP) as a zero‑code bridge that lets large language models seamlessly access databases, external APIs, and execute code, detailing its benefits for developers and everyday users, its core components, step‑by‑step workflow, real‑world examples, and how it outperforms traditional APIs in modern AI agent systems.
This article introduces the fundamentals of augmented large language model agents, explains six workflow patterns—including chain, parallel, routing, orchestrator‑workers, evaluator‑optimizer, and autonomous agents—and provides complete Spring Boot 3 code examples, configuration, and test results for each pattern.
The article analyzes why context engineering is crucial for multi‑agent AI systems, illustrates the fragility caused by fragmented context with a Flappy Bird analogy, and proposes three detailed speculative components—a compression‑to‑structure pipeline, a hybrid layered memory architecture, and a context‑aware coordination mechanism—culminating in a unified reference design for long‑horizon agents.
This article explores the evolving concept of AI agents, debates their definitions, outlines four major deployment challenges—including prompt instability, planning balance, domain knowledge integration, and response speed—and presents practical strategies such as prompt engineering, workflow design, multi‑agent architectures, and model optimization to build reliable, high‑availability agents.
This article explains how constrained decoding, built on context‑free grammars, Jinja templates, and the XGrammar engine, can enforce strict SQL syntax and custom business rules during LLM generation, enabling reliable, production‑grade NL‑to‑SQL services.
The article introduces Model Context Protocol (MCP), explains its architecture of Host, Client, and Server, describes its components—Resources, Tools, Prompts—and demonstrates practical integration with IDE plugins to extend LLM capabilities such as real‑time ticket queries, highlighting its significance for AI development.
This article walks through the challenges of traditional traffic replay, explains the design and benefits of a conventional replay system, and then details how integrating large language models can automate data preparation, script generation, and validation to make backend testing more accurate, scalable, and efficient.
This article examines the LiveCodeBench Pro benchmark, revealing that while large language models achieve impressive scores on knowledge‑ and logic‑heavy coding problems, they still fall short of human experts on high‑difficulty, observation‑intensive tasks, especially without external tool support.
Large language models are booming, and this article highlights six cutting‑edge research directions—LLM‑plus synthetic data, reward modeling, inference techniques, LLM‑as‑a‑Judge, safety alignment, and long‑context handling—each illustrated with recent papers, experimental results, and links to code repositories.
This article presents SCALE, a community‑driven, open‑source benchmark that expands beyond simple Text‑to‑SQL accuracy to evaluate large language models on performance, dialect conversion, and deep SQL understanding, offering developers, researchers, and CTOs a realistic measure of AI‑assisted database tasks.
This comprehensive guide walks you through the fundamentals of Retrieval‑Augmented Generation (RAG) and AI agents, explains their inner workings, shares optimization tricks, provides ready‑to‑run code snippets, and demonstrates how to evaluate performance with metrics such as recall, faithfulness, and answer relevance.
This article examines the theory and practice of shared memory in multi‑agent systems, tracing its evolution from classic blackboard models to modern solutions like Mem0.ai, Open Memory, and A‑MEM, and provides concrete design patterns, integration strategies, and future research directions for LangGraph users.
The article details MicroOps' virtual network feature on Kubernetes, covering manual and intent‑driven deployment, topology visualization and editing, node types, monitoring with Prometheus and Fluentd, chaos injection via ChaosMesh and VN_Chaos, and upcoming alarm and self‑healing modules.
This article presents a step‑by‑step guide for constructing a scalable enterprise Retrieval‑Augmented Generation (RAG) solution using the Model Context Protocol (MCP), covering architecture comparison, system design, Milvus‑backed knowledge store, Python client implementation, deployment scripts, code examples, and best‑practice recommendations.
The article analyzes how large language models can acquire true reasoning abilities for hard‑to‑score industry tasks by combining Chain‑of‑Thought prompting with reinforcement learning, addressing vague reward signals, reward hacking, and loyalty, and proposing a toolbox of reward engineering, synthetic data, hierarchical RL and multi‑agent collaboration.
EasyDistill, an open-source framework from Alibaba Cloud PAI, streamlines knowledge distillation for large language models, introducing the DistilQwen-ThoughtX series with variable-length chain-of-thought reasoning, and provides comprehensive best-practice guidance for training, fine-tuning, evaluation, compression, and deployment via the PAI-ModelGallery.
This guide walks through using Alibaba's new Qwen3-Embedding and Qwen3-Reranker models to build a two‑stage Retrieval‑Augmented Generation pipeline with Milvus, covering environment setup, data ingestion, vector indexing, reranking, and LLM‑driven answer generation, demonstrating production‑grade performance across multilingual queries.
This article chronicles the evolution from simple chat interactions to sophisticated autonomous agents, detailing stages of LLM development, ReAct reasoning, memory management, tool integration, and practical implementation using the browser-use project, while offering prompt design insights and future directions for AI agents.
Eino is an open‑source Golang framework for building large‑model applications, offering reusable components, robust orchestration, clean APIs, best‑practice templates, and full‑cycle DevOps tools, with code examples for both Ollama and OpenAI modes, plus streaming and normal output options.
This article explains how the Model Context Protocol (MCP) standardizes AI agent communication with external tools, outlines its benefits, describes its core components, showcases open‑source implementations, and provides step‑by‑step Python examples for building MCP servers and clients.
This article traces the evolution of AI application architectures—from the earliest minimal user‑LLM interaction to advanced designs featuring context enhancement, input/output guardrails, intent routing, model gateways, caching strategies, agent capabilities, monitoring, and inference performance optimizations—providing practical insights and references for developers.
This comprehensive guide explores the core implementation paths for large language model (LLM) applications, focusing on agent design, workflow orchestration, tool integration, memory management, multi‑agent architectures, and future trends, providing actionable methodologies and real‑world examples for practitioners.
The article shares practical experiences using large language models such as Cursor and DeepSeek in data‑warehouse workflows, covering assisted coding, automated metric extraction, self‑service analysis, documentation generation, their benefits, limitations, and the broader impact on data engineering roles.
The article enumerates common pitfalls of Retrieval‑Augmented Generation—such as missing content, low‑rank document misses, context limits, format errors, incomplete answers, scalability bottlenecks, complex PDF extraction, data‑quality issues, domain adaptation gaps, hallucinations, and feedback‑loop deficiencies—and offers concrete mitigation strategies ranging from data cleaning and prompt design to hybrid search, hierarchical retrieval, document compression, and automated evaluation.
This article examines the challenges of manual data extraction in data‑driven enterprises, explains why large language models alone fall short, and details how the Cursor‑Agent framework automates end‑to‑end querying, knowledge‑base integration, and result validation to become a self‑sufficient "data master" for both technical and non‑technical users.
This article examines how the rise of large language models reshapes core business assets, defines agents and tools, explores multi‑agent collaboration patterns, task allocation and conflict resolution mechanisms, and evaluates the MCP protocol and engineering requirements for building scalable, flexible agent platforms.
Amazon’s newly open‑source Strands Agents SDK lets developers create AI agents with minimal code by defining prompts, tools, and models, offering a lightweight, production‑ready framework that supports multiple model providers, observability, multi‑agent collaboration, and extensible tooling via dedicated packages.
This article surveys the evolution of AI applications, explains large language model fundamentals, outlines architectural challenges, and introduces three core patterns—Retrieval‑Augmented Generation (RAG), autonomous Agents, and Model Context Protocol (MCP)—while providing practical LangChain code snippets and integration guidance.
This paper introduces the Thought Leap Bridge task and the CoT‑Bridge model, which detect and fill missing intermediate steps in chain‑of‑thought reasoning, dramatically improving large language model performance on mathematical and logical benchmarks and enhancing downstream distillation and reinforcement‑learning pipelines.
The article explains that hallucinations in large language models stem mainly from the supervised fine‑tuning stage, illustrates the issue with concrete examples, and presents mitigation techniques such as knowledge‑probing data generation and web‑search tool integration using special tokens.
This article examines the evolving role of the AI engineer, contrasting it with AI researchers, ML engineers, and software engineers, outlines essential skills such as prompt engineering, MLOps, and data integration, and predicts how AI engineering will become a pivotal, high‑demand discipline in the coming years.
This article critically examines seven high‑profile reinforcement‑learning papers for large language models, exposing flawed baseline evaluations, unrealistic settings, and modest actual improvements despite bold claims of dramatic performance gains.
MarkerGen introduces a novel, plug‑and‑play framework that decomposes length‑controllable text generation into four sub‑abilities—identifying, counting, planning, and aligning—integrates external tokenizers and dynamic markers, and achieves significantly lower length errors and higher quality across diverse models, tasks, and languages.
This article provides a comprehensive overview of large language models (LLMs), covering their transformer architecture, parameter counts, GPU memory and storage requirements, and detailed fine‑tuning methods such as prompt engineering, data construction, LoRA, PEFT, RLHF, and DPO, along with practical deployment and inference acceleration strategies.
This article outlines essential LLM engineering skills, including scripts for converting various model checkpoints to Llama format, customizing modeling files for advanced features, building a multi‑GPU inference class, and adding channel‑aware loss tracking to fine‑tuning pipelines.
Discover how to design and implement an MCP Server that integrates AI-driven observability, covering essential components, best practices, code examples, and real-world lessons learned to enable natural language interaction with monitoring data and streamline system analysis.
This article provides a systematic overview of existing LLM AI Agent communication protocols, categorizing them into four major types, detailing their functions, benefits, and use‑cases, and compares four representative protocols—MCP, A2A, ANP, and Agora—through a concrete travel‑planning scenario.
This article provides a comprehensive technical analysis of LLM context, covering its definition, types, tokenization, window‑size evolution, diminishing returns, management techniques such as RAG, CoT, memory‑as‑a‑service, and future challenges like multimodal fusion, privacy, and autonomous agent memory.
The author recounts fixing a complex Redis Vector Sets bug, explores how human creativity outperforms LLMs in devising efficient data‑consistency checks, and shares experimental ideas—including XOR accumulators and MurmurHash—to detect non‑mutual links in large HNSW graphs.
This guide explains how to construct a deep research workflow in Dify that leverages AI agents, loop variables, and structured outputs to automatically explore complex topics, gather sources, and synthesize comprehensive reports with proper citations.
This article explains how the Model Context Protocol (MCP) standardizes AI‑assistant communication, compares the traditional Server‑Sent Events (SSE) transport with the newer Streamable HTTP mechanism, and provides step‑by‑step code examples for building both MCP servers and clients that leverage Streamable HTTP for bidirectional, session‑aware data exchange.
This tutorial walks through creating a simple bigram language model in pure Python, refactoring it into a PyTorch implementation, and explains core concepts such as tokenization, embedding layers, loss functions, gradient descent, training loops, and text generation, preparing you for building a full GPT model.
This article introduces the OmniThought dataset, which annotates over two million chain‑of‑thought reasoning steps with Reasoning Verbosity and Cognitive Difficulty scores, and explains how these metrics guide the training of DistilQwen‑ThoughtX models that adapt chain length to task difficulty, achieving superior performance compared to existing distilled LLMs.
This article provides a concise, easy-to-understand introduction to large language models, the transformer architecture, prompt engineering, temperature settings, function calling, the Model Context Protocol (MCP), agent communication (A2A), and future AI programming trends, using simple analogies and illustrative examples.
This article provides a comprehensive guide to large language model fine‑tuning, covering model architecture, parameter and memory calculations, prompt engineering, data construction, LoRA and PEFT techniques, reinforcement learning methods such as DPO, and practical deployment workflows on internal platforms.
This article explains why Human-in-the-Loop (HITL) is essential for enterprise LLM agents, outlines common HITL patterns, and shows how LangGraph’s interrupt, resume, and checkpoint mechanisms can be used to build reliable, auditable workflows with tool‑call control and practical code examples.
Version 0.8.0 of Dify introduces parallel workflow capabilities, allowing multiple branches to run concurrently, which dramatically reduces latency for complex LLM tasks; the guide explains how to create simple, nested, iterative, and conditional parallel branches, with step‑by‑step instructions and visual examples.
This comprehensive guide explains the Text2SQL concept, showcases real‑world scenarios, compares three implementation architectures—including a simple prompt‑based method, a LangChain‑based pipeline, and an enterprise‑grade Vanna solution—while providing practical tips, security measures, and advanced enhancements for deploying robust natural‑language‑to‑SQL systems.
This article compares two recent papers that investigate why large reasoning models such as Llama and Qwen show degraded instruction‑following performance when using chain‑of‑thought prompting, analyzing attention patterns, training effects, and proposed mitigation strategies.
This article compares three leading AI workflow tools—n8n, Dify, and Coze—by examining their origins, technical architectures, core advantages, typical use cases, real‑world case studies, and future deployment trends, helping developers and businesses choose the right "intelligent assistant" for their needs.
A recent study from Harvard, Amazon and NYU shows that using chain‑of‑thought (CoT) prompting can significantly reduce large language models' ability to follow strict instructions, introducing a new "constraint attention" metric and four mitigation strategies to restore performance.
This article details the end‑to‑end design, parsing tricks, vector database setup, retrieval strategies, prompt engineering, and LLM reranking that powered the winning solution in a company‑annual‑report question‑answering competition.
This article explains how a SaaS company leveraged large language models to automatically ingest, enrich, and analyze stability alerts, turning noisy notifications into actionable insights through configurable pipelines, Feishu integration, and a streamlined AI workflow that boosts incident response speed and reduces manual effort.
This tutorial walks you through creating a FastMCP‑based weather‑query server in Python, registering it as an LLM‑callable tool, and building a matching Python client that connects via stdio, handles tool calls, and provides an interactive chat loop for AI‑driven queries.
This article introduces TickIt, a ByteBrain system that leverages large language models to automatically identify and escalate critical Oncall tickets, detailing its multi‑class escalation, deduplication, and category‑guided fine‑tuning modules, experimental results, and the operational impact on cloud services.
The article recounts Xiaoting’s journey from a PhD research background to leading JD.com’s ad‑fraud detection, detailing how large language models, reinforcement learning, and model distillation were applied to identify hidden address codes, reduce false‑positive rates to 0.3%, and balance accuracy with real‑time performance in a high‑traffic e‑commerce environment.
This article recounts a JD researcher’s journey from academic data‑mining competitions to building an AI‑driven, LLM‑enhanced anti‑fraud platform that balances detection accuracy, computational cost, and business value in large‑scale e‑commerce advertising.
The article examines the inherent knowledge cutoff of large language models, introduces function calling, Model Context Protocol (MCP), and Agent‑to‑Agent (A2A) as solutions for real‑time data access, compares their architectures, communication patterns, and use cases, and discusses their respective strengths and drawbacks.
This article explains the challenges of integrating diverse tools without MCP, then demonstrates step‑by‑step how to configure Spring‑AI and the native MCP SDK to call LLMs, register tools, handle SSE and stdio services, and troubleshoot common issues, providing code snippets and best‑practice recommendations.
This article presents a comprehensive AI-driven solution for automatically generating functional test cases, detailing the AI background, design scheme, core components such as PRD parsing, test‑point generation, test‑case creation, knowledge‑base construction, implementation results, and future development directions.
AI MCP (Model Context Protocol) is an open protocol that standardizes communication between large language model applications and external data sources or tools, offering pre‑installed services, fast registration, ecosystem openness, and automatic discovery within Huawei Cloud ModelArts Studio, while eliminating the need for per‑API integration code.
This article describes how a hybrid architecture that combines third‑party address‑recognition APIs with large‑language‑model (LLM) processing, along with carefully engineered prompts and a TSV output format, dramatically improves address parsing accuracy and latency in a retail checkout scenario.
The article explains that large language models lose up to 35% performance in multi‑turn conversations, critiques static single‑turn evaluation methods, and proposes a dynamic user‑simulator with loss‑masking techniques to generate realistic test turns and improve assessment reliability.
This article explains the Model Context Protocol (MCP), its components, and how to integrate MCP servers with Alibaba Cloud Observability 2.0, offering practical design experiences, tool simplification tips, default parameter strategies, output size control, and future AI‑driven observability insights.
This article explains why residual connections are essential in deep neural networks, describing the problems of network degradation and gradient vanishing, how shortcut paths add the input to the layer output, the requirement of matching dimensions, and the resulting stability for training large language models.
This article demystifies AI agents and agentic workflows, explaining their core components—LLMs, tools, and memory—while detailing planning, tool‑use, and reflection patterns, comparing agentic, non‑agentic, and traditional workflows, and exploring real‑world applications, advantages, and limitations.
This guide walks you through creating a ChatBI intelligent agent by integrating Alibaba's Qwen3 large language model with PAI‑LangStudio, configuring the Model Context Protocol (MCP) server, and connecting to Hologres real‑time data warehouse, covering setup, deployment, and verification steps for enterprise data analysis.
LangChain4j 1.0.0 brings official OpenAI SDK support, GitHub Models integration, expanded database and vector store compatibility, customizable HTTP clients, and clear migration steps for renamed interfaces and streaming methods, marking a major milestone for Java AI development.
This article presents StableReinforce and the R1-Reward model, demonstrating how reinforcement learning techniques can stabilize training and significantly improve the performance of multimodal reward models for large language models across several benchmarks.
StarRocks MCP Server provides a universal adapter that lets large language models like Claude, OpenAI, and Gemini execute SQL queries directly against StarRocks, simplifying data Q&A, intelligent analysis, and automated reporting by eliminating the need for bespoke plugins or complex prompt engineering.
The article reviews the shortcomings of traditional large language models, explains how function calling extends LLMs beyond pure text, introduces the Model Context Protocol (MCP) as a standardized USB‑C‑like interface for AI tools, and demonstrates a Python MCP example that integrates LLMs with Tencent Advertising APIs.
This article provides a detailed Chinese‑to‑English summary of Andrej Karpathy’s 7‑hour LLM tutorial, covering chat process analysis, tokenization, pre‑training data pipelines, model architecture, training strategies, post‑training fine‑tuning, reinforcement learning, chain‑of‑thought reasoning, and current industry applications.
The article introduces Chain‑of‑Recursive‑Thoughts (CoRT), explains how recursive self‑evaluation enhances large language model reasoning, outlines its workflow, shares GitHub resources, compares it with existing CoT methods, and reports experimental results using Mistral 3.1 24B.
This article walks through Amap's Model Context Protocol (MCP) service, explaining its purpose, installation steps, configuration in the Cursor client, and practical examples such as travel planning and location‑based queries, while also evaluating its strengths and current limitations.
This article outlines Anthropic's yearly insights on constructing large‑language‑model agents, explaining their definitions, when to employ them, recommended frameworks, modular building blocks, common workflow patterns, and real‑world application scenarios for developers.
This article explains the Mixture of Experts (MoE) technique used in modern LLMs, detailing its core components—experts and router—comparing dense and sparse layers, describing load‑balancing, expert capacity, and routing strategies, and showcasing real‑world examples such as Switch Transformer, Vision‑MoE, and Mixtral 8x7B.
The article examines how pre‑generating structured context for codebases can overcome the uncertainty and quality issues of traditional Retrieval‑Augmented Generation, outlines the technical and business challenges of RAG, compares existing code‑search tools, and introduces AutoDev’s Context Worker as a practical solution.
This article explains the design of an AI gateway that centralizes LLM access, detailing its background, overall architecture, authentication, quota management, multi‑model routing, load‑balancing strategies, multi‑tenant isolation, observability features, and the supported API protocols for enterprise integration.
