YR

Loading Experience

Yuvraj Raina

Software Engineer focused on building polished, high-performance web experiences.

Warming up the scene

Open Source

Small tools with a bias for local control, clear APIs, and practical deployment.

This collection brings together lightweight graphics components, identity verification models, local AI tools, and self-hosted creative software, each shaped around a narrow problem and a usable developer surface.

Cross-platform reach

200K+ views

Clones and installs

3,000+

Public repos

05

React Three Fiber

TinyEarth

react-earth-lite

A tiny, realistic Earth object for React Three Fiber with compressed NASA-derived maps and a restrained runtime footprint.

A production-friendly globe primitive with sensible defaults and no extra scene machinery.

TinyEarth is built for projects that need a polished globe without inheriting a full globe framework. It keeps the API focused, supports clouds, night lights, atmosphere, custom texture paths, markers, and quality presets, while keeping the visual cost predictable.

<10 KB gzipped target

2 default draw calls

<20 MB GPU texture target

What it handles

Compressed Blue Marble-style mapsOptional clouds and atmosphereInstanced markersReduced-motion aware
ReactThree.jsReact Three FiberWebP textures
GitHubtinyearth.yuvrajraina.com

AI Face Verification

TinyFaceMatch

tinyfacematch

An open-source face verification model that compares two aligned face images and returns a match decision with similarity scores.

Small enough to ship, explicit enough to evaluate, careful enough for sensitive identity workflows.

TinyFaceMatch focuses on practical verification: a 128-D embedding model, ONNX runtime support, a documented threshold, Python usage, embedding comparison helpers, and production notes around alignment, monitoring, consent, and biometric data handling.

99.72% reported accuracy

0.9983 ROC AUC

13.238 MB FP32 model

What it handles

Aligned face comparisonL2-normalized embeddingsSimilarity and distance outputDjango API backend path
PythonONNXNumPyDjango
GitHubtinyfacematch.yuvrajraina.com

Passive Anti-Spoofing

TinyLiveness

tinyliveness

A lightweight RGB face liveness prototype for detecting whether a face crop appears live rather than a photo, video, or screen spoof.

A practical anti-spoofing building block for identity flows that need more than face similarity.

TinyLiveness is the liveness companion to TinyFaceMatch. It ships a default ONNX model and policy, Python and JavaScript runtime paths, reproducible evaluation notes, and clear caveats about larger holdouts and production camera testing.

98.25% reported accuracy

0.999325 ROC AUC

15.296 MB FP32 ONNX

What it handles

Live / spoof / manual review decisions224px RGB face crop contractPython and JavaScript runtimesDocumented APCER policy
PythonONNX RuntimeJavaScriptEfficientNet-B0
GitHubtinyliveness.yuvrajraina.com

Local Image Generation

ChitraDev

A local-first FastAPI image generation app for Stable Diffusion 1.5 with optional FaceID reference guidance.

A self-hosted creative workstation for image generation without hosted quotas or remote credits.

ChitraDev keeps generation local: prompts, reference photos, generated PNGs, ZIP downloads, and model cache stay on the user machine. It supports prompt-only runs, single-person FaceID, two-person FaceID, placement controls, and batch generation for longer local jobs.

Local FastAPI UI

Stable Diffusion 1.5

FaceID guidance

What it handles

Prompt-only generationOne or two FaceID referencesBatch creatorLocal output and cache control
FastAPIStable DiffusionIP-Adapter FaceIDPython
GitHubLocal app

Local AI Developer Tooling

LocalPilot

A privacy-first VS Code coding assistant powered by local Ollama models for chat, inline suggestions, fixes, explanations, and tests.

An AI coding loop that keeps code on the developer machine by default.

LocalPilot brings Copilot-style workflows to local models: inline ghost text, a chat sidebar, selected-code commands, code actions, error explanations, model setup, low-RAM modes, and a safe apply flow with diff preview before edits land.

Local Ollama calls

No telemetry

Low-RAM modes

What it handles

Inline code suggestionsChat over selected codeFixes and generated testsSafe diff preview workflow
VS Code ExtensionTypeScriptOllamaLocal LLMs
GitHublocalpilot.yuvrajraina.com