Build a fully local voice-to-code pipeline using Whisper and Claude Code

·1 views

A developer on DEV Community has detailed a fully local voice-to-code setup that keeps all data — audio, context, and source code — on the user's machine without sending anything to the cloud. The pipeline combines faster-whisper for speech transcription and Claude Code's built-in /voice command, introduced in March 2026, as the coding agent. Audio is captured via sox, transcribed locally using the Whisper medium model, and fed directly into the agent's prompt. Benchmarks on a MacBook M2 show a transcription latency of roughly 1.2 seconds per phrase with the medium int8 model, compared to 200–450 milliseconds for cloud-based Codex voice. The guide is aimed at developers handling client code or sensitive projects where cloud-based voice tools like ChatGPT or Codex pose a data privacy concern.

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 Immersive Scroll Narrative Landing Page for Fictional Indonesian Restaurant

A developer has completely redesigned a fictional Indonesian comfort food restaurant landing page called Rasa Kenangan, meaning 'Flavor of Memories', set in Yogyakarta. The project was built using only vanilla HTML, CSS, and JavaScript with no frameworks or backend, and is deployed statically on Cloudflare Pages. The redesign replaced a generic hero-and-grid layout with an immersive scroll-driven experience featuring parallax animations, sticky storytelling sections, and a draggable horizontal menu. Design inspiration was drawn from Awwwards-winning restaurant websites such as BALANS KITCHEN and CAÍN, with a warm amber and charcoal color palette using modern oklch color space. The finished page achieved a perfect Lighthouse score of 100 across performance, accessibility, best practices, and SEO.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How Angular Developers Can Build Smart Objects Using NgRx SignalStore Enrichment

A technical guide published on DEV Community explains how Angular developers can move beyond passive data containers by transforming raw API responses into Smart Objects using a Factory Mixin pattern called Object Enrichment. The approach involves extending plain TypeScript data interfaces with domain methods such as fullName() and hasRole(), making business logic self-contained and immediately discoverable via IDE autocomplete. The article argues that enriching data at write time—when it enters the NgRx SignalStore—is architecturally safer than enriching it at read time via computed() signals, since the latter still exposes raw state to components that bypass the enriched layer. Angular's Dependency Injection constraints within the store are addressed using the runInInjectionContext utility. A working implementation is available on both GitHub and StackBlitz for developers to explore directly.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Dart Developers Can Now Build ADK Agents and MCP Servers Without Python or Node.js

A new open-source project demonstrates how Dart developers can build AI agents and Model Context Protocol (MCP) servers using the unofficial adk_dart library and the Shelf HTTP framework, without relying on Python or Node.js. The repository, adk-hello-world-dart, includes two examples: a CLI agent using ADK's LlmAgent with a registered function tool, and a Shelf server exposing an MCP-compatible greeting tool over Server-Sent Events. The server handles JSON-RPC routing through a SessionService component and supports core MCP methods such as initialize, tools/list, and tools/call. The project targets Dart 3.5 or later and is intended as a practical workaround, as no official Dart ADK had been released as of July 2026. The author notes that the transport and JSON-RPC implementation are intentionally minimal and not designed as a general-purpose MCP server.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

GTM Engineering Is Systems Design — and Most Teams Are Building It Wrong

Go-to-market (GTM) engineering, a role that barely existed two years ago, is now a staple in growth team hiring plans, with US salaries commonly ranging from $110,000 to $180,000. At its core, the role involves building automated pipelines that convert raw prospect data into sales conversations with minimal human intervention, using tools like Clay, n8n, and increasingly LLM agents. According to Daria, a 12-year developer-tools GTM veteran and founder of GTM Labs, the standard high-volume outreach architecture fails when targeting developer audiences, who are technically savvy, distrust marketing language, and readily identify templated messages. She argues that effective GTM systems for technical buyers should prioritize high-quality behavioral signals over contact volume, and distribute trusted artifacts like benchmarks and documentation rather than automated email sequences. Developers considering a career pivot into GTM engineering hold a natural advantage, she notes, because they already understand how these systems work and can judge which automations are genuinely useful.

0 comments Read more at DEV Community