Skip to main content
3Nsofts logo3Nsofts
Research & Development

Lab

I prototype the future. Experimental projects exploring AI-native workflows, local-first architectures, and unconventional system designs. These are R&D efforts—learning in public.

Active Experiments

Current research projects and proof-of-concept work.

Nick — Behavioral Threat Scoring Engine

Live · v1.0

On-device CoreML threat correlation for macOS

An open-source behavioural threat scoring engine built into Nick. Six independent monitors feed a 30-second sliding ThreatCorrelator, which outputs a 0–1 score via CoreML. Foundation Models generates plain-English alert explanations — all on-device, zero cloud dependency.

What we built

  • •ThreatCorrelator: 30-second sliding window across six monitor signals → CoreML 0–1 behavioural score
  • •libyara + CoreML hybrid scanner — signature-based and behavioural YARA heuristics
  • •Foundation Models on-device LLM generates zero-cloud alert explanations in plain English
  • •Interactive threat scoring simulator — try 6 preset attack scenarios and 24 manual signal toggles
Swift 6CoreMLFoundation ModelsAGPL-3.0
Product page →

AI-Native OS Experiments

Active R&D

Rethinking desktop workflows with local AI

Exploring what happens when the operating system is designed around conversational AI, not file hierarchies. Local LLM integration, natural language command translation, and privacy-first architecture experiments.

What I'm testing

  • •Shell interfaces that accept natural language and translate to system commands
  • •Context-aware file management using local LLMs (Ollama, llama.cpp)
  • •Battery-efficient inference strategies for continuous AI assistance
  • •Zero-telemetry, zero-cloud AI workflows for complete privacy

Current status

Early prototypes demonstrating feasibility. Command translation works reliably for common tasks. Exploring UI paradigms that blend traditional desktop metaphors with conversational interaction.

PythonOllamaLocal LLMsShell
Full findings →Product page

Local-First Data Sync Patterns

Research

Exploring CRDT and event sourcing architectures

Investigating patterns for production-grade local-first applications that work offline, sync across devices, and handle conflicts gracefully without custom backend infrastructure.

What I'm learning

  • •CloudKit limitations and when to build custom CRDT implementations
  • •Event sourcing for audit trails in business applications
  • •Hybrid architectures: when to use Apple frameworks vs custom solutions
Core DataCloudKitCRDTsEvent Sourcing
Full findings →

On-Device ML Optimization

Experimental

Battery-aware inference strategies for mobile

Testing quantization techniques, model pruning, and runtime optimization strategies to make on-device AI practical for battery-constrained environments like iOS and watchOS.

Focus areas

  • •Core ML model optimization and quantization workflows
  • •Apple Neural Engine performance characteristics and trade-offs
  • •Adaptive inference: adjusting model complexity based on battery state
Core MLApple Neural EngineQuantization
Full findings →

Research Philosophy

The Lab is where I test ideas without the constraint of immediate production viability. Some experiments inform client work. Others become products. Many teach us what not to build.

I share findings publicly when possible—documenting what worked, what didn't, and the technical trade-offs discovered along the way.

This isn't about innovation theater. It's about maintaining technical edge through continuous exploration of emerging patterns, tools, and architectures.

Interested in Experimental Work?

If you're exploring similar problems or want to collaborate on forward-looking projects, let's talk. Some Lab experiments evolve into productized offerings or client engagements.

Get in touchView services