You are
now
viewing
Scroll
to
explore
Research &
engineering
as one discipline
© 2026
Lightyear 360 Labs
Pvt Ltd
AI Lab
You are
now
viewing
Scroll
to
explore
Research &
engineering
as one discipline
Copyright ©2026
Lightyear 360 Labs
Pvt Ltd
Model Families
Foundation models and fine-tuned models we build and ship in-house.
SLM360™
On-device AI that thinks, learns, and remembers. 50MB, 39ms latency, 100% offline.
Med360™
Medical AI models for clinical transcription, diagnosis, and multilingual healthcare. Med360 Lite + Sanad-1.0.
The Harness
The agent harness we build our models and tools on.
Platforms & Tools
Foundation models, languages, and infrastructure we build and ship in-house.
Aura
AI-native language and compiler for product UI. One compact source file compiles to native SwiftUI, Jetpack Compose, and HTML/CSS/JS with design tokens and security types built into the grammar.
Omni 1.0™ & OmniScient™
Two 360 Labs foundation models on public leaderboards: enterprise text-to-SQL (Spider 2.0-Lite) and enterprise time-series forecasting (TIME benchmark).
Crawl360
Production-grade web scraping API with auto-escalating fetcher modes, structured data extraction, multi-page crawling, screenshot capture, PDF generation, and AEO360 SEO audits.
Mem360
Universal AI memory engine that extracts, stores, and retrieves long-term user memories for any AI application. Pluggable LLM, storage, and embedding layers.
Research
Read our technical papers on Med360 and SLM360 architecture, training methodology, and benchmarks.
Med360: A Family of Fine-tuned Multilingual Medical AI Assistants for Indian Healthcare
We present Med360, a family of fine-tuned large language models designed specifically for Indian healthcare contexts with native Hinglish support, Indian pharmaceutical nomenclature, and AIIMS/NEET-PG level medical knowledge.
SLM360: Breaking the NLU Trilemma - Accuracy, Latency, and Efficiency On-Device
A lightweight NLU engine achieving 98-100% accuracy at 39ms latency in 50MB -4x faster than Rasa, 10x less memory. Hybrid classification with reasoning, memory, and prediction, all on-device.
AgentGuard: Deadlock Prevention for Multi-AI-Agent Systems via Extended Banker’s Algorithm
A C++17 library that extends Dijkstra’s Banker’s Algorithm for LLM agent orchestration, with progress monitoring, authority cycle detection, and adaptive demand estimation. 285 tests, LangGraph integration.
VAJRA: A Multi-Sensor On-Device Counter-UAS System with Custom-Trained Visual and Acoustic Deep Learning Models
A fully on-device, multi-sensor drone detection and neutralization system running on a commercial Android smartphone. Integrates visual (YOLOv8n), acoustic (FFT + CNN), and RF analysis -fused into a unified threat display with countermeasure control. 38MB APK, zero network dependency.
KAVACH: An On-Device Tactical Intelligence Platform with SLM360-Powered Natural Language C2, Real-Time ISR, and Automated Reporting
A tactical intelligence platform running entirely on Android with zero network dependency. Integrates Tactical C2, ISR Processing (YOLOv8n, ~19 FPS), Auto SITREP (SALUTE reports in ~8s), and Patrol Optimization -all powered by SLM360 (577K params, 848KB).
Blog
Technical deep-dives and engineering insights.
Your LangGraph Agents Are Deadlocking. Here’s the Fix.
How we applied a 60-year-old OS algorithm to prevent deadlocks in multi-AI-agent systems. A practical guide to using AgentGuard with LangGraph.
Quick Start
Get started with Pi360 in just a few lines of code.
1# Install Pi3602curl -fsSL https://pi360.omnun.ai/install | sh3# Launch the agent4pi360