AR.Labs
Advanced Research Laboratory
AI Engineering Through a Cognitive Science Lens
We build voice agents, agentic systems, and cognitive architectures that augment human capability.
Research Focus
Currently Building
Container diagram for Sales Bot Voice Agent
Level 2: System ContainersAcademic Foundation
Exeter University · Carnegie Mellon University
BSc (Hons.) Cognitive Science — Published Research
Production Systems
4,000+ lines production Kotlin
Coherence: Voice-First AI Coach
Research Focus
11 research projects, 10+ tools & experiments
Voice AI & Agentic Systems
Projects
Production systems and active engineering work
Sales Bot Voice Agent
Production Voice Sales Agent
Full-stack voice sales agent using LangGraph state machines for SPIN-selling conversation flow. Real-time voice via Pipecat, with React frontend and FastAPI backend. Currently reverse-engineering and deploying.
Key Achievements
- +5-phase SPIN selling conversation state machine
- +3-app architecture: React ↔ FastAPI ↔ Pipecat voice pipeline
- +858 tests across the full stack
- +Real-time bidirectional voice with STT/TTS
VOX Second Brain
Cognitive AI Infrastructure
AI-powered personal infrastructure system with progressive disclosure architecture, context engineering, and multi-agent coordination. The system building this website.
Key Achievements
- +30+ custom skills and autonomous workflows
- +Progressive disclosure 4-layer cognitive hierarchy
- +CEO Agent for strategic cadence management
- +Voice-first interface with real-time TTS
Multi-Agent Orchestration
AI Agents Building Software Together
Coordination system where multiple AI agents collaborate on software engineering. Agent Mail for messaging, shared workspace via filesystem, beads for task tracking. Three agents built the Sales Bot in 4 days.
Key Achievements
- +3 AI agents built Sales Bot (26K lines) in 4 days
- +Agent Mail: inter-agent messaging via MCP
- +Shared workspace with filesystem coordination
- +Task tracking across agents via beads
Active_Projects
CEO Agent
Strategic AI orchestrator with cadence state machine
VOX Scheduler
Voice-first calendar with domain/mode system
Coherence
Voice-first AI coaching Android app
Building in Public
Content strategy and publishing system
Job Search Pipeline
Systematic BD with PRP methodology
Graph Memory Research
Episodic memory for AI agents
Richard M. Thompson
Software & AI Engineer-in-Training
20 years ago, I studied Cognitive Science at the University of Exeter (BSc. (Hons.) First Class), then interned at Carnegie Mellon University with Prof. J.L. McClelland — one of the founders of connectionism. We published research on underlying cognitive structures of children's reasoning about causality, specifically examining learning and memory using an early, partly hand-coded recurrent neural network.
I spent the years between then and now pursuing different career paths. Business consulting, coaching, education, video game testing, personal training, and carpentry.
I'm now applying what I've learnt in these careers to software and AI engineering, at a most interesting juncture for technology.
I'm most interested in voice AI, agentic architecture, and cognitive approaches to AI system design.
Latest_Transmissions
CORE-LOOP: Managing Complex Two-Way Conversations via Agent/LLM
Deep dive into 4-chain dialogue architecture for conversational agents
Adaptive Cognitive Architecture for HCI Design
How cognitive science principles inform AI system design
AI-Driven Android Voice Assistant Concepts
From concept to production: building voice-first mobile AI
Capabilities
[SYSTEM_CAPABILITIES_MANIFEST_V1.0]
Work With AR Labs
AI Engineering // Voice Agents // Agentic Systems