[360Labs.ai]
{ 360Labs }Case Study
{ Research and Development }· Live

SLM360

On-device NLU engine with 98-100% accuracy, 39ms latency, 50MB footprint. Solves the NLU trilemma.

SLM360
// overview

Hybrid classification: fast pattern matching (<1ms) + semantic embeddings (39ms) with confidence arbitration. Multi-step reasoning, 5-tier SmartMemory, predictive context, entirely on-device in 50MB.

// specs

Specifications

Accuracy100% Banking77, 98% SNIPS
Latency39ms P50, <1ms pattern-only
Footprint50MB total system
Memory5-Tier SmartMemory (~64KB/user)
// features

Features

  • 01Fast Path: compiled regex + keywords (<1ms)
  • 02Semantic Path: quantized 32MB ONNX model, 384 dims (~35ms)
  • 03Arbitration layer selects best result
  • 045-Tier SmartMemory: Short-term, Episodic, Semantic, Procedural, Meta-learning
  • 05Beats Rasa (~96%) and Dialogflow (~97%) on SNIPS at 98%
// tech stack

Tech Stack

TypeScriptONNX RuntimeWebAssemblyNode.js

Interested
in SLM360?

Let's discuss how this can work for you.

Get in Touch →
// end of case studySLM360
Contact