Developer Builds AI Workout Timer That Yells Coaching Cues in Real Time

·2 views

A developer built an AI-powered workout timer called Sonic Kinetic that generates fully custom exercise routines and spoken coaching audio on demand. Users input a callsign, available time, target muscle groups, and desired intensity, after which Google's Gemini model creates a structured interval timeline with a coaching line for each segment. ElevenLabs then converts every coaching line into audio concurrently using Go goroutines, keeping generation time short regardless of routine length. The app also adjusts the ElevenLabs voice stability setting to a lower value for maximum-intensity segments, making those cues sound more intense. The Angular-based frontend displays a real-time pacing curve as an SVG and highlights the active interval during playback, with the full project source available on GitHub.

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)

Virtual Thread Pinning: How a Silent JVM Bug Caps Your App's Throughput

Java's virtual threads, introduced via Project Loom, can silently revert to behaving like a bounded thread pool due to a phenomenon called 'pinning,' where a virtual thread cannot unmount from its carrier OS thread during a blocking operation. This occurs in exactly two scenarios: blocking inside a synchronized block (in JDK 21–23) and blocking within native or JNI frames, the latter remaining unfixed even in JDK 24. When pinning occurs repeatedly in a hot code path, all carrier threads — limited by default to the number of CPU cores — can stall simultaneously, causing scheduler starvation that mimics a deadlock from the outside. JDK 24 resolves the synchronized pinning issue by decoupling object monitors from carrier threads, but native frame pinning persists and can hide in unexpected places like static initializers. Developers are advised to audit blocking calls inside synchronized methods and third-party libraries, and use JVM diagnostics to detect pinning before it silently degrades throughput.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer cuts AI costs by routing agent calls through coding assistants instead of LLM APIs

A developer has shared a method of replacing direct large language model API calls with coding assistant agents, such as Claude Code or GitHub Copilot, to reduce costs when building AI systems. Standard LLM API pricing typically ranges from $2 to $7 per million tokens, while coding assistant subscriptions can translate to as little as $0.08 per million tokens. The approach involves wrapping a coding assistant's command-line interface in a lightweight integration layer, enabling it to function as a drop-in LLM backend without a native API. A Python module using subprocess calls to the Claude CLI was developed to support both plain-text and structured outputs, including a LangChain-compatible adapter. The key trade-off is that coding assistants lack a direct API, but the author argues this is easily overcome through CLI-based wrappers.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Deploying a Solana Program to Mainnet: Key Risks and Responsibilities

Moving a Solana program from devnet to mainnet is a significant shift that goes beyond running a few commands. On mainnet, deployment costs real SOL, the program is immediately accessible to any user or bad actor, and any mistakes are permanently recorded on a public ledger. A critical decision developers must make before launch is whether to retain or revoke the program's upgrade authority, which controls the ability to push future updates. Keeping upgrade authority allows bug fixes but requires users to trust the developer will not push malicious changes, while revoking it makes the program immutable and maximally trustworthy but eliminates any ability to patch vulnerabilities. Most teams treat this as a nuanced, staged decision rather than a simple binary choice made at launch.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Slack Debuts AI-Driven Agentic Testing; Devs Weigh Memory Layers for TypeScript Agents

This week's developer-focused roundup covers three key areas of AI agent development: memory architecture, test automation, and content workflows. Slack Engineering has introduced 'Agentic Testing,' an AI-driven approach where agents autonomously explore user interfaces to identify bugs and edge cases, replacing brittle hand-written test scripts. On the memory front, a technical comparison examines Mem0 and TurboMem, two solutions for managing state in TypeScript-based AI agents, weighing service-based versus embedded memory architectures. A separate piece traces one developer's evolution from simple prompt files to unified agent-driven content automation workflows. Together, these articles reflect growing industry interest in applying AI agent orchestration to practical, production-grade engineering challenges.

0 comments Read more at DEV Community