BUILDING HIGH-PERFORMANCE LOW-LEVEL SYSTEMS.
CS @ Chennai Institute of Technology · Engineering sub-1 GFLOP Edge AI models, low-level Linux kernel interception, and offline agent runtimes.
Knight Badge (Top 1.5% globally)
CGPA at Chennai Institute of Technology
Ti Forge, AI for Sustainability, ServiceNow
30+ FPS INT8 Real-Time Drone Model
SUB-1 GFLOP VISION ENGINE
INT8 TensorRT quantization layer active across drone camera pipeline with 96% mAP.
SYSTEM CONFIG
FLAGSHIP SOFTWARE SHOWCASE.
Test and run production systems engineered for low-level performance, zero-leakage local AI agents, and developer toolchains.
Local AI Agent SDK & Runtime
Loopback-only local AI agent runtime & typed Python streaming SDK.
A private, local-first LLM orchestration system that connects your terminal, local workspace index, and host telemetry through a lightweight Rust launcher and typed Python SDK without sending data to third-party clouds.
Strict loopback only
@jarvis sub-ms latency
Async streaming client
100% offline isolation
$git clone --depth 1 https://github.com/ravindran-dev/localmind.git "$HOME/.local/share/localmind" && bash "$HOME/.local/share/localmind/install.sh"
Loopback-only local AI agent runtime, terminal assistant, and Python streaming SDK.
System-level evdev remapping, shortcut conflict detection, and typing practice.
Your music library, kept close and under your control.
A focused, offline-first coding workspace for Windows.
An Arch Linux developer desktop, from Sway to shell and editor.
One package-management interface across Linux distributions.
Start a focused Windows workspace; restore the system when you finish.
Private, peer-to-peer file sharing for nearby Android devices.
ENGINEERING CAPABILITIES.
LOCALMIND: PRIVATE AGENT RUNTIME
Loopback-only local AI agent runtime & typed Python streaming SDK. Binds to 127.0.0.1:8420 with sub-ms Rust launcher and zero cloud leakage.
from localmind_sdk import LocalMindClient
client = LocalMindClient()
async for event in client.stream_agent("profile host"):
print(event.token, end="")KEYNIX: INPUT PLATFORM
Hardware-level Linux input interception using evdev EVIOCGRAB exclusive grabbing and /dev/uinput virtual HID device synthesis.
SUB-1 GFLOP DRONE AI
Real-time drone object detection models engineered under 1 GFLOP compute, achieving 96% mAP with TensorRT INT8 quantization.
HERTIX: LOCAL-FIRST MUSIC PLAYER
Cross-platform music player & media library engineered with native Android (Kotlin, Jetpack Compose, Media3 background audio) and Windows (React 19, Tauri 2, Rust) with synchronized lyrics.
HOW I ENGINEER SYSTEMS.
From hardware-level kernel interrupts to edge AI quantization and typed developer toolchains.
KERNEL & HARDWARE VIRTUALIZATION
Direct manipulation of hardware events via Linux evdev exclusive grabbing (EVIOCGRAB), eBPF sandboxing, and /dev/uinput virtual HID synthesis. Bypasses user-space lag to guarantee deterministic <0.1ms latency.
EDGE AI QUANTIZATION & AGENT RUNTIMES
Quantizing neural vision pipelines into sub-1 GFLOP TensorRT INT8 engines achieving 60+ FPS on embedded drone chips. Deploying private LLM orchestrators locked to loopback sockets (127.0.0.1:8420) with zero cloud data leakage.
PRODUCTION TOOLCHAINS & TYPED SDKS
Packaging systems with native Rust launchers (@jarvis), typed Python event streaming SDKs, and sub-second SAT package solvers. Backed by SHA-256 cryptographic package validation and rootless rollbacks.
LET'S BUILD SOMETHING RELENTLESS.
Ready to deploy low-level systems, edge AI inference pipelines, and mission-critical software.