The article details how Anthropic's Claude Code secretly gathers Chinese users' timezone and proxy information by embedding hidden markers in system prompts through steganography, explains the detection logic, discusses the privacy implications, and notes Anthropic's plan to remove the code in an upcoming release.
OpenHuman is an open‑source AI assistant written in Rust that runs locally on a laptop, offers zero‑cloud data storage, integrates 118+ services via OAuth, uses a Memory Tree for persistent context, provides SuperContext zero‑wait prompts, and includes TokenJuice compression to cut token costs up to 80%.
Microsoft’s OneDrive sent a security influencer a bundle of untouched private photos from 2014 via its “Memories” feature, sparking a wave of criticism that highlights how cloud‑storage algorithms can unintentionally expose sensitive data years after users think it’s deleted.
After finding commercial remote tools like Sunflower unreliable, the author evaluates the open‑source RustDesk, explains its self‑hosted architecture, walks through server and client installation on Linux, and shows how it delivers stable, private remote desktop access.
iOS 27 quietly integrates AI into eight everyday iPhone functions—from bill splitting and password updates to smarter messaging, call context, calendar parsing, shortcut automation, home notifications, and Safari tab grouping—showcasing Apple’s philosophy of invisible, privacy‑first intelligence.
Even with meticulous cleaning, residual noise, redundant legal clauses, and approximate duplicates can degrade retrieval and generation in RAG systems, while privacy risks from embedding inversion and the need for continuous, metric‑driven governance make data quality the ultimate ceiling for performance.
This survey systematically reviews trustworthy agentic AI, focusing on safety and robustness as well as privacy and system security, mapping risks and safeguards across the agent lifecycle, proposing unified metrics and benchmarks, and discussing high‑risk real‑world applications and open challenges.
Anthropic announced that, starting July 8, Claude users will need to upload government ID and complete a live facial scan via Persona, a move that threatens free‑tier accounts with stricter bans, raises data‑security concerns, and excludes mainland Chinese IDs, while API developers remain unaffected.
type-review is a privacy‑first, browser‑only typing trainer that adaptively unlocks letters, offers benchmark testing, detailed per‑key statistics, optional visual keyboards and sound packs, and is built with SolidJS, TypeScript and a strict three‑layer architecture.
The article analyzes four post‑exam security threats—platform data leaks, phishing attacks, fake hacker‑grade‑change scams, and privacy‑risk social‑media posts—and provides concrete steps for students and parents to protect personal information.
WWDC26 put Siri AI at the forefront of iOS 27, turning AI from a peripheral feature into a system‑level assistant, while sparking debate over Apple’s reliance on external cloud models, privacy implications, and the long‑term risk of mixing rented compute with its tightly integrated ecosystem.
The article examines why emerging wearables, smart glasses, and in‑car systems need a "first‑layer" on‑device AI that preprocesses multimodal inputs, outlines the missing input, application, and system capabilities required for edge models, and discusses how subsequent edge and cloud stages should share the workload.
Open Notebook offers a fully open‑source, locally deployed AI note‑taking platform that prioritizes data privacy, supports over 18 AI providers, provides multimodal content handling, customizable podcast generation, and extensible REST APIs, positioning it as a comprehensive, privacy‑enhanced alternative to Google Notebook LM.
Supermemory, the open‑source AI memory engine, eliminates repeated forgetting by offering a zero‑configuration, multi‑modal memory layer that tops LongMemEval, LoCoMo and ConvoMo benchmarks, integrates automatic learning, mixed RAG‑Memory search, built‑in connectors, privacy tags, and multiple deployment options from no‑code web to local offline versions.
Headroom is a locally run, reversible context‑compression layer for Claude Code that reduces input tokens by 60‑95 % without sacrificing precision, eliminates context‑limit errors, cuts token costs, protects privacy, and enables seamless memory sharing across multiple AI coding agents, as demonstrated by real‑world benchmarks.
The article explains how, even with encrypted payloads, the timing, size, and direction of network traffic generated by LLM agents can be fingerprinted to reveal user behavior and long‑term profiles, posing significant privacy threats beyond content protection.
When Discord switched to a teen‑by‑default policy, users discovered that a simple thumb sketch with two eyes and a mouth could trick the on‑device AI age estimator into granting adult access, exposing the limits of lightweight facial analysis models.
The author evaluates Marvis, an AI‑powered desktop assistant that bundles six specialized agents—fast terminal scheduling, autonomous planning, cross‑modal task chains, a visual agent workspace, vibecoding for code, and desktop organization—showcasing rapid local execution, privacy‑preserving design, multi‑agent coordination, and future mobile integration.
In May 2026 a hacker named NormalLeVrai released roughly 3 billion WhatsApp records on the dark web, prompting a Texas lawsuit against Meta, a public accusation by Telegram’s Pavel Durov, and a detailed technical analysis exposing gaps between WhatsApp’s end‑to‑end encryption theory and its real‑world implementation, followed by risk assessments and mitigation advice for enterprises and individuals.
A technical review reveals that Telegram’s MTProto protocol exposes a permanent 64‑bit device identifier (auth_key_id) in clear text, enabling passive observers—including ISPs, mobile carriers, and state surveillance—to track users across app restarts, IP changes, VPNs, and even Tor, rendering secret chats and PFS ineffective.
AnySearch launches a unified API that aggregates 22 professional data sources for AI agents, using intent classification and RRF fusion to cut token usage by up to 70% and boost accuracy and latency over Parallel and Brave, while offering architecture‑level privacy protections.
The article outlines five critical setup steps—privacy with direnv and a secret manager, token handling via litellm or portkey, context management using uv and git commits, visibility through mitmproxy, and rigorous evaluation with inspect‑ai—showing how they cut token waste by 68.3%, reduce costs 92.5% and raise evaluation pass rates to 94.2% across 347 runs.
The article analyzes the long‑standing privacy dilemma of cloud‑based agents, presents MemPrivacy’s three‑stage de‑identification framework and four‑level privacy taxonomy, details its two‑phase training with the MemPrivacy‑Bench dataset, and shows benchmark results where a 0.6B model outperforms GPT‑5.2 while keeping latency under 0.5 seconds.
MemPrivacy introduces a reversible, fine‑grained privacy layer for edge‑cloud agents, outperforming OpenAI's privacy‑filter by over 50 % F1 while keeping system utility loss under 2 %, thus enabling agents to remain useful without exposing raw sensitive data.
OpenHuman tackles the common AI‑assistant problems of slow cold‑start, complex integration, and weak privacy by offering a minimalist desktop UI, over 118 built‑in service integrations, local memory trees with Obsidian compatibility, and a self‑developed TokenJuice compression that cuts token usage by up to 80 %, all under a GNU open‑source license.
The article critically examines ten major problems of modern AI, from excessive energy and water consumption and dubious employment promises to pervasive surveillance, militarization, ethical abuses, copyright theft, mental‑health risks, societal erosion, job displacement, and the looming AI hype bubble.
Chrome version 147 silently downloads a 4 GB Gemini Nano model for optional on‑device AI features, which many users never use, and the article explains the privacy concerns and provides step‑by‑step methods to disable the download via flags or Windows registry.
Local Deep Research is an open‑source AI assistant that runs entirely on a consumer RTX 3090, reaches about 95% accuracy on the SimpleQA benchmark, uses a plugin‑based architecture with multiple LLM and search back‑ends, stores data in an encrypted SQLCipher database, and can be launched in minutes via Docker for privacy‑focused researchers and developers.
Tencent’s Hy‑MT1.5‑1.8B‑1.25bit model, now open‑source, runs entirely offline on smartphones, supports 33 languages, and—according to internal tests—delivers translation quality that surpasses Google Translate’s online service, highlighting the impact of 1.25‑bit quantization on model size and performance.
With CI/CD pipelines maturing, automated test coverage is no longer the bottleneck; the real constraint has shifted to producing accurate, fast, and secure test data, prompting teams to upgrade from simple data mocking to full‑stack data engineering, AI‑driven synthesis, and verifiable data contracts.
Apple’s recent AI paper probes whether large vision‑language models truly forget user data by examining residual streams and final logits, revealing that hidden image attributes persist in top‑k outputs and exposing significant privacy and security risks.
This article provides an in‑depth, eight‑dimension comparison of Hermes Agent and OpenClaw, examining their core philosophies, learning abilities, integration options, deployment ease, security, standout features, overall strengths, and guidance on selecting the right AI agent for different user scenarios.
This practical guide compares Hermes Agent and OpenClaw across user skill levels, privacy requirements, and automation goals, offering step‑by‑step installation tips, scenario‑specific recommendations, and a decision matrix to help you choose the most suitable open‑source AI agent.
The paper reveals that intermediate representations in vertical federated learning retain exploitable geometric structure, and introduces VENOM—a geometry‑aware model‑stealing framework that outperforms existing defenses across multiple datasets, even under distribution shift.
The article details a step‑by‑step journey of creating a private, continuously evolving AI knowledge base—from single‑file markdown archives to modular Skills, data sanitization, Git‑based version control, and automated daily curation—showing why richer personal data and closed‑loop feedback are essential for a truly useful AI assistant.
A detailed security analysis shows that LinkedIn's browser extension silently runs a heavily obfuscated JavaScript payload that probes over 6,200 Chrome extensions in a few milliseconds, building precise user profiles, political tags, and commercial intelligence, while Microsoft defends the practice as anti‑scraping.
In a landmark forensic case, the FBI extracted deleted Signal messages from an iPhone by accessing the system‑level notification database, exposing how Signal’s disappearing‑message feature can be bypassed, detailing the underlying iOS storage flaw, community reactions, and practical steps users can take to mitigate the risk.
An explosive GitHub project called ‘Colleague‑Skill’ lets you feed a former coworker’s chat logs, documents, and personal habits into an AI, creating a digital replica that mimics their coding style, tone, and decision‑making, sparking debates on workplace automation, privacy, and the future of digital labor.
Citizen Lab’s investigation reveals that the Webloc platform harvests advertising IDs and real‑time bidding data from billions of mobile devices, enabling law‑enforcement agencies in the US, EU and elsewhere to track half a billion users worldwide, while most users remain unaware of the privacy risks.
Google’s AI Edge Gallery lets you run cutting‑edge large language models such as Gemma 4 entirely offline on Android or iOS devices, offering absolute privacy, zero‑latency responses, and a modular platform with agent skills, thinking mode, multimodal input, and a prompt‑lab for on‑device AI experimentation.
An in‑depth examination of the leaked Claude Code repository shows how Anthropic’s AI client can retain extensive user data, execute privileged commands, and embed covert telemetry, raising serious security and privacy concerns for both public and private deployments.
This article provides a detailed comparative analysis of nine domestic Claw AI agents in 2026, evaluating their pricing models, core functionalities, and ease of use, and offers concrete recommendations for both personal and enterprise users while highlighting common pitfalls.
The article breaks down the 2026 ShinyHunters breach list by industry, revealing that Salesforce systems were the primary target, and details massive data exposures across finance, government, retail, and social platforms while urging victims to secure accounts and avoid downloading the leaked files.
The article presents a privacy‑first system that extracts numeric telemetry from each LLM interaction, discards raw text, and evaluates whether detection of prompt injection and jailbreak attacks remains effective, showing only a 1.4 F1‑point drop when using solely text‑independent features.
OpenClaw is an open‑source, self‑hosted AI assistant that lets you keep all data locally while chatting through familiar tools like Telegram and Discord, switch freely among Claude, GPT and Gemini models, extend functionality with a rich toolset, and control devices via a node system, all orchestrated by a central Gateway.
The recent one‑click WeChat integration of OpenClaw signals a shift of AI agents from niche developer tools to the everyday chat window, lowering activation cost, reshaping user‑agent relationships, and raising new privacy and trust challenges for product strategy.
Violoop introduces a plug‑and‑play hardware runtime that fills the missing execution layer for AI agents by combining visual perception, system API signals, and direct HID control, enabling autonomous perception‑judgment‑execution loops, secure dual‑chip permissions, cost‑effective edge inference, and 24/7 scheduling without relying on fragile RPA scripts.
The article provides a comprehensive analysis of core voice interaction system capabilities—including barge‑in, turn‑taking, multi‑turn dialogue, intent recognition, speaker identification, streaming latency, noise robustness, multilingual support, emotion handling, personalization, security, and deployment considerations—highlighting typical scenarios such as smart speakers, in‑car assistants, call centers, and meeting transcription.
A leading identity verification provider exposed over one billion personal records after a cloud storage bucket was misconfigured, revealing names, IDs, biometric data and more; the breach impacted finance, e‑commerce, government and social platforms, prompting analysis of technical and managerial failures and a set of remediation steps for individuals, enterprises and the industry.
This article introduces DeerFlow 2.0, Ruflo v3.5, and claude‑relay‑service, detailing their core features, deployment options, performance benefits, and GitHub repositories, and offers practical recommendations for individual developers and enterprises seeking production‑grade Claude agent solutions.
The article examines the massive public queue for free OpenClaw installations at Tencent’s campus, explains the tool’s local‑run AI agent architecture, analyzes market forces driving its rapid adoption, and critically assesses the technical, security, and business challenges that temper its long‑term prospects.
A recent paper from Sapienza University's GLADIA Lab shows that mainstream Transformer language models are injective, enabling a novel SIPIT algorithm to recover original text from hidden states with perfect accuracy, while extensive experiments confirm the models retain all input information.
OpenClaw, a self‑hosted AI assistant gateway, surged past React and Linux on GitHub, amassing over 250,000 stars in just four months, and the article analyzes its explosive growth, underlying factors, security concerns, operational costs, and what this shift means for the meaning of star counts.
Meta is secretly testing an AI shopping assistant that leverages billions of users' social profiles, shifting recommendation logic from reactive behavior data to proactive identity‑driven suggestions, while raising significant privacy and ecosystem implications for e‑commerce.
OpenClaw shattered GitHub star records by amassing 245,000 stars in two months—surpassing React’s 12‑year tally—thanks to a privacy‑first AI assistant gateway, ultra‑low setup friction, a booming Skills plugin ecosystem, and rapid developer‑driven viral spread, despite rising costs and security challenges.
A recent U.S. court case revealed that Microsoft can hand over BitLocker recovery keys to law enforcement, allowing the FBI to bypass Windows encryption and access a suspect's laptop, prompting a discussion of the privacy trade‑offs and steps users can take to regain control of their keys.
OpenClaw is an open‑source, headless AI agent that runs locally, interacts through WhatsApp, Telegram or Discord, stores persistent memory, and can automate tasks ranging from car‑price negotiation to autonomous phone calls, while raising significant security, privacy, and operational challenges.
A UCL computer‑science team presents a federated learning framework for white‑blood‑cell morphology analysis that preserves patient privacy, leverages heterogeneous clinical slide data from multiple sites, and achieves superior cross‑site performance and generalisation to unseen institutions compared with centralized training.
This article explains why front‑end recording and replay are essential for diagnosing invisible bugs, introduces the open‑source rrweb tool and its core modules, compares it with video and screenshot methods, and provides step‑by‑step guidance for installing, configuring, recording, replaying, and optimizing rrweb in Vue 3 applications.
Frustrated by limited, paid online converters, the author discovers the open‑source ConvertX tool, which supports over a thousand formats, runs locally via a single Docker command, and offers privacy‑first, ad‑free file conversion for both single and batch tasks.
Apple will unveil a Gemini‑powered Siri in late February, a strategic shift that lets the voice assistant access personal data, understand screen content, and perform complex tasks, while raising privacy questions and signaling a broader industry move toward AI collaborations.
This guide walks you through installing Ollama on Ubuntu, loading the open‑source DeepSeek‑R1 model, configuring AnythingLLM as a local RAG system, and testing it offline so the AI can answer questions from your private documents without any data ever leaving your machine.
AgentCPM-Report is a locally deployable, privacy‑preserving AI agent that matches or exceeds the performance of top closed‑source large‑model systems on deep‑research benchmarks, offering end‑to‑end report generation without uploading any confidential data to the cloud.
A developer‑maintained PowerShell script hosted on GitHub removes built‑in AI components from Windows 11, aiming to improve privacy and security, while sparking broader debate about AI integration, ethical concerns, and the effectiveness of AI‑assisted development.
The first national medical AI data security and privacy standard, backed by the National Health Commission and industry leaders, outlines a full‑lifecycle risk‑control framework and calls on hospitals, AI firms, regulators and legal experts to join its drafting process.
brow6el is a terminal‑only web browser that leverages the Sixel bitmap format and the Chromium embedded framework to deliver fully rendered pages, mouse input, bookmarks, download management, ad‑blocking, and Vim‑style navigation, offering a privacy‑focused alternative to AI‑laden mainstream browsers.
A comprehensive dialogue among industry experts explores the concept of memory engineering for AI agents, covering its definition, system‑level challenges from edge to cloud, hybrid technical routes, evaluation metrics, privacy safeguards, audience questions, future directions, and practical advice for developers.
Open Notebook is a fully local, open‑source AI notebook that rivals Google Notebook LM by supporting over 16 LLM providers, handling multimodal content, and enabling advanced multi‑speaker podcast generation while giving users complete data sovereignty and flexible deployment options.
This guide explains how to replace common cloud services—such as storage, document editing, video streaming, ad blocking, and password management—with open‑source, self‑hosted alternatives that run on Docker or a Raspberry Pi, offering full data control, no ads, unlimited bandwidth, and low‑cost deployment.
Sam Altman and Alex Blania's Worldcoin project, which aims to create a global human‑identity infrastructure with iris‑scanning Orb devices, has attracted only about 1.75 million verified users—roughly 2% of its billion‑user target—while facing regulatory bans in the Philippines and scrutiny in multiple other countries over privacy and data‑handling concerns.
The article evaluates Doubao’s AI phone by testing everyday scenarios, highlighting its screen‑recognition‑driven automation, high latency, privacy risks, and comparing its performance and usability against Honor’s YOYO assistant and other AI‑enabled smartphones.
The article analyzes how AI is reshaping education by identifying five product value breakthroughs, outlining five implementation pitfalls, and envisioning future AI‑driven learning experiences, emphasizing a shift from mere tool substitution to a symbiotic, human‑centered value model.
LocalAI is an open‑source, self‑hosted alternative to OpenAI that lets you run large language, image and audio models on your own CPU or GPU, offering full data privacy, zero cloud costs, and offline capability while remaining compatible with the OpenAI API ecosystem.
This article explains how to protect sensitive e‑commerce chatbot logs by using Alibaba Cloud Log Service (SLS) masking functions together with LoongCollector to collect, mask, and store logs securely, enabling operations, product, and security teams to analyze data without exposing personal information.
This weekly roundup spotlights three open‑source tools—AFFiNE’s local‑first knowledge workspace, Vanna’s natural‑language‑to‑SQL AI framework, and Handy’s offline, privacy‑focused speech‑to‑text app—showcasing recent advances in knowledge management, data analysis, and secure voice transcription.
The article reveals that Microsoft's Gaming Copilot feature captures on‑screen text via OCR and uploads it to the cloud for AI model training, discusses privacy concerns, performance impacts on games like Battlefield 6, and provides steps to disable or uninstall the feature.
This roundup covers recent security flaws in Google Pixel phones, subtle async Rust pitfalls called Futurelock, SQLite concurrency insights, Debian's Rust mandate, ICE facial‑recognition concerns, open‑source Rubik's cube automation, chaotic attractor visualizations, arXiv policy changes, and the EU chat‑control proposal's defeat.
A recent breach of the AI companion apps Chattee and GiMe Chat exposed over 40 million private conversations, hundreds of thousands of media files, and transaction records due to publicly accessible Kafka brokers lacking any authentication, putting user privacy and security at serious risk.
This article explains the origins, technical classification, and evolution of the IP address 1.1.1.1 from an unassigned block to Cloudflare's fast, privacy‑focused public DNS, covering its network segment, Anycast deployment, encrypted DNS support, and comparison with other DNS services.
At the ACM CCS 2025 live paper showcase, three groundbreaking studies—Gibbon’s fast secure two‑party GBDT training, Panther’s efficient private approximate nearest‑neighbor search on a single server, and PromeFuzz’s knowledge‑driven LLM approach to fuzzing harness generation—are presented, highlighting significant performance and security advances.
The article explains how Claude Code’s decision to use real‑time grep instead of code indexing reflects a 50‑year‑old Unix philosophy, showing that stateless design improves composability, scalability, predictability, and privacy across AI assistants, serverless platforms, and distributed systems.
This article explains data masking (data desensitization), its importance for privacy and compliance, outlines the four implementation stages, compares common masking techniques, and discusses the challenges and future directions for secure data handling in development and testing.
The article explains what a VPN is, clarifies its legal status in various countries, debunks common myths about privacy, describes how VPNs act as encrypted proxies requiring trust in providers, and offers guidance on selecting a reliable service or building your own.
Atlassian’s $6.1 billion cash acquisition of The Browser Company aims to embed AI into its suite, sparking a debate on the viability of AI‑enhanced browsers like Dia, Perplexity Comet, Edge Copilot and Brave Leo, and exploring their potential to reshape productivity, privacy, and the future of web interaction.
The article examines how major platforms like WeChat, Bilibili, Apple, Facebook, and Blizzard handle users' digital assets after death, highlighting policies such as account reclamation, memorial accounts, and inheritance features, while noting the scarcity of profit-driven solutions for digital legacy preservation.
This article presents a curated list of ten open‑source AI tools—from instant prototyping agents and reactive notebooks to fast LLM fine‑tuning, ethical hacking assistants, local ChatGPT interfaces, and database‑integrated machine learning—explaining their key features, benefits, and why developers should adopt them to boost productivity and maintain privacy.
A new study reveals that large language models which excel on static medical exams dramatically lose accuracy when subjected to the Dynamic, Automatic, Systematic (DAS) red‑team framework, exposing serious weaknesses in robustness, privacy, bias, and hallucination, and urging the adoption of continuous adversarial evaluation for trustworthy clinical AI.
This article explains how Cursor creates Merkle‑tree hashes for change detection, uses Tree‑sitter for syntax‑aware code chunking, generates vector embeddings stored in Turbopuffer, and employs privacy‑preserving mechanisms to enable fast, secure code‑base search and autocomplete.
Sam Altman cautions against over‑reliance on ChatGPT, while Germany blocks DeepSeek for GDPR violations, Tencent unveils its MoE‑based Hunyuan‑A13B model, and Google releases a Python client for Data Commons, highlighting both AI risks and rapid open‑source advancements.
The HTTP Referer header, a ubiquitous but often overlooked request field, records the source page of a user, aids traffic analysis, enables anti‑hotlinking and CSRF protection, and carries a historic misspelling that led to the Referrer‑Policy standards governing privacy and security.
The article analyses how privacy concerns, data‑security priorities and an application‑first strategy cause both Apple’s Apple Intelligence and WeChat’s AI features to lag behind hype, examining product decisions, technical constraints, and the potential future of AI agents within these ecosystems.
This article presents six practical data‑desensitization methods—including string replacement, encryption, database masking, cache‑based replacement, dynamic AOP masking, and K‑anonymity—explains their principles, shows Java implementations, compares security, performance, and reversibility, and offers concrete recommendations for protecting sensitive fields such as phone numbers and ID cards.
This article presents six widely used data masking solutions—from simple regex string replacement to advanced K‑anonymity—detailing their principles, Java implementations, pros and cons, performance impact, and suitable application scenarios, helping developers protect sensitive data in production and test environments.
The article presents the abstracts, publication types, links, and research areas of seventeen Ant Group papers accepted at ICLR 2025, covering topics such as embodied robot co‑design, efficient distributed training for large language models, optimization via LLMs, character animation, interactive frame interpolation, KV‑cache management, and privacy‑preserving Transformers.
This case study details how 58租房 tackled cumbersome landlord publishing by redesigning the workflow with smart location (LBS), AI‑driven shooting assistance, OCR‑based document recognition, and digital‑human guidance, achieving up to 90% faster operations, higher accuracy, and stronger privacy protection.
This article presents concise English summaries of the 18 Ant Group papers accepted at AAAI 2025, covering topics such as privacy‑preserving large‑model tuning, knowledge‑graph integration, AI‑generated image detection, multi‑task learning, generative retrieval, role‑playing evaluation, and video hallucination mitigation.
This article compares the traditional Xposed‑based dynamic privacy checker with a new Gradle‑plugin instrumentation approach, explains the design and implementation of the Transform API‑driven solution, shows performance metrics, provides usage instructions, and outlines automated testing and future improvements for Android apps.
This article examines the rapid growth of AI agents, outlines their core components and classifications, analyzes a wide range of privacy and security threats—including data leakage, prompt injection, jailbreak, backdoor, hallucination, and memory attacks—and proposes practical governance measures to mitigate these risks.
This article presents a curated overview of fifteen Ant Group research papers accepted at NeurIPS 2024, covering topics such as large language models, knowledge graphs, recommendation systems, privacy-preserving inference, and multimodal learning, with abstracts, paper types, links, and key contributions highlighted.
This article describes a privacy‑preserving, low‑cost logistics QA bot that combines data cleaning, augmentation, BM25 and vector retrieval, a DNN rerank model, and LLM‑based answer rewriting, along with evaluation results and deployment considerations.
