Welcome to The Weekly Five - your curated list of 5
exceptional open source projects I discovered this week.
The Weekly Five: Open Source Tools That Reimagine Infrastructure
Infrastructure used to mean servers and databases. Now it means sensing with WiFi signals, autonomous ML engineers, and decentralized app stores that bypass corporate gatekeepers. This week's projects push the boundaries of what we consider foundational technology, from camera-free spatial intelligence to agents that read research papers and ship models. Each one rethinks how we interact with compute, data, and distribution.
Top takeaways
WiFi becomes a sensor: RuView transforms commodity WiFi into a spatial intelligence platform that detects presence, tracks movement, and monitors vital signs without cameras or wearables.
Autonomous ML engineering: ml-intern reads papers, writes code, trains models, and ships them to Hugging Face with minimal human intervention.
Decentralized distribution: komi-store reimagines app installation by turning GitHub, Codeberg, and Forgejo releases into a one-click app store experience.
Who this issue is for
Developers and infrastructure engineers interested in privacy-preserving sensing, agentic ML workflows, or alternatives to centralized app distribution.
RuView
Why this made the cut: Camera-free spatial sensing with commodity hardware turns every WiFi router into a privacy-preserving presence detection and vital sign monitoring system.
Why it matters
RuView proves that WiFi Channel State Information (CSI) can replace cameras for many spatial intelligence tasks. It detects people through walls, measures breathing and heart rate, estimates human pose (via DensePose), and monitors rooms in complete darkness. No cameras, no wearables, just physics and signal processing. This opens privacy-preserving use cases for home automation, elderly care, and occupancy sensing where video would be invasive or impractical.
Key features
DensePose estimation from WiFi signals without any visual input
Vital sign monitoring including breathing rate and heart rate detection through RF analysis
ESP32 firmware for deploying on low-cost IoT hardware
Home Assistant integration for automating smart home workflows based on presence and activity
Self-learning algorithms that adapt to specific environments and improve accuracy over time
How to use
Start with an ESP32 board or any WiFi device that exposes CSI data. Extract CSI from WiFi packets, preprocess the signal data, and run pose estimation models. For home automation, integrate RuView with Home Assistant to trigger scenes based on room occupancy or detected activity. When setting up, flash the ESP32 firmware and calibrate for your specific room layout. Community experience shows that strategic access point placement maximizes signal coverage, and running initial calibration sessions trains the system on your environment for better accuracy.
🔗 View on GitHub | GitHub stars: 73,914
ml-intern
Why this made the cut: An open-source autonomous agent that handles the full ML lifecycle (research, coding, training, deployment) using the Hugging Face ecosystem.
Why it matters
ml-intern compresses the ML engineering workflow into a single agent. It reads research papers, explores Hugging Face docs, writes training code, launches compute jobs, evaluates results, and publishes models to the Hub. This autonomy is especially valuable for rapid prototyping, fine-tuning experiments, and post-training workflows like RLHF or DPO where iteration speed matters. It lowers the barrier for teams without dedicated ML engineers and accelerates research-to-production cycles.
Key features
Autonomous research that reads papers and Hugging Face documentation to inform model design decisions
End-to-end execution from writing training scripts to managing datasets, launching cloud compute, and shipping models
Hugging Face integration with deep access to the Hub, Transformers library, datasets, and Spaces for deployment
Local model support via provider-specific model prefixes instead of commercial APIs
Multi-step workflows that chain tasks like data preprocessing, hyperparameter tuning, and evaluation automatically
How to use
Install ml-intern via pip, configure API keys for Hugging Face (and optionally cloud compute providers), then prompt it with high-level goals like "fine-tune a sentiment classifier on this dataset and publish to the Hub." The agent generates training code, monitors the run, and pushes the final model. For LLM post-training automation, set up RLHF pipelines by providing the base model and preference dataset. Community recommendations suggest starting with smaller models and datasets to validate the agent's code before scaling to production runs, and using server mode to select local models for cost-sensitive workflows.
🔗 View on GitHub | GitHub stars: 10,457
komi-store
Why this made the cut: A decentralized, open-source app store that bypasses traditional distribution platforms by installing directly from GitHub, Codeberg, and Forgejo releases.
Why it matters
komi-store (formerly GitHub Store) challenges the app store duopoly by treating git forges as first-class distribution platforms. Developers publish releases on GitHub, Codeberg, or Forgejo, and users discover and install apps with one click. This model eliminates gatekeepers, review delays, and platform fees. It is especially relevant for open-source developers who want direct relationships with users and for users who prefer transparency and control over what they install.
Key features
Multi-platform support for browsing releases from GitHub, Codeberg, and Forgejo in a unified interface
One-click installation that automatically downloads, verifies, and installs apps from release artifacts
Cross-platform on Android, Linux, macOS, and Windows using Kotlin Multiplatform
Open by default with all source code, discovery mechanisms, and installation logic auditable
Decentralized discovery with no central registry or approval process, just direct forge integration
How to use
Download komi-store for your platform from the latest releases page. On first launch, the app indexes popular repositories and displays available apps. Search by name, topic, or language, then click to install directly from the forge's release artifacts. The app handles extraction, permission setup, and desktop integration automatically. For developers, publishing to komi-store simply means tagging releases on GitHub, Codeberg, or Forgejo with standard release binaries. No additional registration or submission process required.
🔗 View on GitHub | GitHub stars: 15,343
cedar
Why this made the cut: A React and GraphQL framework designed to simplify full-stack web app development with strong typing and Prisma integration.
Why it matters
cedar reduces boilerplate in React + GraphQL applications by providing conventions for schema-first development, type-safe data fetching, and Prisma ORM integration. It is particularly useful for teams that want end-to-end TypeScript type safety without manually writing GraphQL resolvers and React hooks. The framework's tight integration with Prisma means database schema changes propagate automatically to GraphQL types and React components.
Key features
GraphQL-first architecture with schema-driven development and automatic resolver generation
Prisma integration that syncs database schema directly to GraphQL types
TypeScript throughout for end-to-end type safety from database to UI components
React optimization with built-in hooks and patterns for efficient data fetching and caching
Hacktoberfest-friendly with an active contributor community and clear pathways for open-source contributions
How to use
Install cedar via npm, scaffold a new project, and define your Prisma schema. Run the code generator to produce GraphQL types and React hooks automatically. Write React components using the generated hooks for type-safe queries and mutations. Deploy to any Node.js host with GraphQL endpoint support. The 479 contributors signal an active community, so check issues and discussions for patterns, plugins, and deployment examples.
🔗 View on GitHub | GitHub stars: 137
strawberry
Why this made the cut: A high-fidelity, open-source music player optimized for audiophiles who demand bit-perfect playback and extensive format support.
Why it matters
Strawberry fills the gap for users who want local music library management with audiophile-grade playback. It supports FLAC, WavPack, and other lossless formats with bit-perfect output via ALSA, eliminating resampling and software mixing that degrade audio quality. Integration with Last.fm and MusicBrainz adds scrobbling and metadata tagging, while the Qt-based interface runs natively on Linux, macOS, and Windows. For anyone dissatisfied with streaming services or lightweight players that sacrifice audio quality, Strawberry delivers control and fidelity.
Key features
Bit-perfect playback with ALSA integration that bypasses PulseAudio/PipeWire for unaltered audio output
Lossless format support for native playback of FLAC, WavPack, and other high-resolution formats
Library organization with automatic tagging, cover art fetching, and MusicBrainz integration
Last.fm scrobbling to track playback history and sync listening data
Cross-platform with consistent experience on Linux, macOS, and Windows via Qt
How to use
Install Strawberry from the official builds page or via package managers on Linux distributions. Point the app at your music library folder and let it scan and organize tracks. Configure ALSA output in settings to enable bit-perfect mode on Linux. Use the MusicBrainz tagger to correct metadata and fetch cover art. For optimal audio quality, disable resampling in system audio settings and use exclusive ALSA device access for the cleanest signal path. The interface supports playlist management, library navigation, and customizable audio settings for different output configurations.
🔗 View on GitHub | GitHub stars: 3,776
If you only try one
RuView is the single best starting point this week. Camera-free spatial sensing with commodity WiFi is both immediately useful (home automation, elderly care, occupancy detection) and conceptually transformative (privacy-preserving sensing without video). The project's guides make setup accessible even if you have never worked with WiFi CSI or ESP32 hardware, and the applications range from practical (presence detection for smart home scenes) to cutting-edge (DensePose estimation through walls). If you care about privacy, home automation, or the future of ambient sensing, start here.
If you are doing Open Source I have a good news for you, I work at CodeRabbit which is an AI review tool and its free for Open Source, please reach out to me on X or LinkedIn or just send an email on [email protected] if you need help on adopting CodeRabbit.
You can visit our portal below to create a new account and connect your repository and start reviewing your code.
