Alibaba's Qwen 3.8 27B Open Model Goes Viral, Runs Locally via Spring Boot

·1 views

Alibaba's Qwen 3.8 27B, a dense open-source vision-language model licensed under Apache 2.0, surged to the top of Hacker News within a day of release, accumulating over 1,194 points and 713 comments. The 27-billion-parameter model supports text, image, and video input with a native context window of 262,144 tokens and runs on consumer hardware such as an M5 Max MacBook Pro using a 17GB GGUF file. Developer Simon Willison tested it locally via LM Studio, noting it produced high-quality output but consumed 22,276 reasoning tokens over 21 minutes for a single task. The model ships with built-in thinking mode, multi-token prediction, and FP8 quantization, and Alibaba claims significant benchmark improvements over its predecessor, Qwen3.6-27B, across coding and computer-use tasks. Developers using Spring Boot and Spring AI can integrate the model locally without additional SDKs or cloud accounts, making it accessible for production AI workflows.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Developer Builds Multilingual English Practice Voice Agent for Indian Users

A developer created SpeakEasy AI, a voice-based AI assistant designed to help Indian users practice English through natural spoken conversations. Built as part of the Murf AI Voice for Bharat Challenge, the tool supports English, Hindi, and Hinglish to accommodate users with varying levels of English confidence. The system uses Python, LiveKit Agents, Murf AI for voice generation, and a FastAPI and Next.js stack, combining speech-to-text, text-to-speech, and LLM-based conversation. A conversational routing feature allows the main agent to redirect users to specialist agents, such as a dedicated maths practice assistant. The project also includes a basic call analytics dashboard and is planned for further development with improved memory, more specialist agents, and expanded multilingual support.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Salesforce Admin Exam Guide: Security Models, Flow Automation, and Data Architecture

The Salesforce Certified Administrator exam tests candidates on translating business requirements into declarative platform architecture rather than rote feature memorisation. Salesforce employs a multi-layered security model spanning organisation, object, field, and record levels, where Organisation-Wide Defaults set the most restrictive baseline and sharing rules can only expand access, never restrict it. With Process Builder and Workflow Rules now deprecated, Salesforce Flow has become the primary automation tool, offering before-save and after-save execution paths suited to different use cases. Schema design knowledge is also critical, particularly distinguishing Master-Detail relationships — which support roll-up summaries and cascade deletions — from loosely coupled Lookup relationships. Exam strategy favours native declarative solutions, elimination of security anti-patterns, and careful time management across 60 questions in 105 minutes.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Graph Theory in C#: BFS, DFS, and Real-World Problem Solving Explained

A technical tutorial published on DEV Community demonstrates how to apply graph theory concepts in C# to solve practical problems such as resolving nested role-based permissions. The article covers core vocabulary, including the distinction between directed and undirected graphs, which the author identifies as the most common source of silent bugs in hand-written graph code. It walks through implementing a generic Graph class using a dictionary of HashSets to prevent duplicate edges and ensure efficient traversal. Breadth-first search (BFS) and depth-first search (DFS) are both explained with working C# code, highlighting key implementation details like when to mark nodes as visited. The guide also addresses finding all connected components by iterating over unvisited nodes and launching fresh traversals from each.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

PawSense AI Uses Google Gemini to Analyze Dog Photos and Generate Canine Profiles

A developer has built PawSense AI, a full-stack web application submitted for DEV Community's Weekend Challenge: Dog Days Edition. The app allows users to upload a dog photo or use a webcam, after which Google Gemini's multimodal vision analyzes the image to produce a detailed canine profile. Features include breed identification, personality archetypes, a simulated 'dog thoughts' translator with text-to-speech, a personalized daily care schedule, and a downloadable ID card called a PawPort. The frontend is built with React 19 and TypeScript, while a Node.js and Express backend handles communication with the Gemini API. The application is live on Google AI Studio and uses IndexedDB for persistent client-side storage of scanned dog profiles.

0 comments Read more at DEV Community