Richard Thompson
AI engineer building production voice agents and autonomous AI systems — cognitive science research foundation, three years of intensive full-stack engineering, and an independent practice delivering AI automation to European and international clients.
Profile
I run AR Labs, an independent AI engineering practice: automation delivery for small businesses, real-time voice systems, and the agent infrastructure that makes AI work dependable.
My path started in research — a First-Class Cognitive Science degree and published neural-network modelling with Prof. J.L. McClelland at Carnegie Mellon — and returned to AI through three years of intensive, full-time software engineering: shipped Android apps, live voice pipelines, and an autonomous software delivery system, all built and documented in public.
The through-line is engineering discipline applied to AI systems: test-driven implementation, independent review gates, and architectures a client can operate after I leave.
Selected Work
Real-Time Voice Sales Agent
Live demoConversational voice agent that talks to website visitors in real time — listens, reasons, and answers with sub-second turnaround.
Streaming voice pipeline built on Pipecat with WebRTC transport, Deepgram speech-to-text, and ElevenLabs voice synthesis, driven by a FastAPI backend and a React front end.
Music School Operations Automation
Client engagementAutomation delivery for an independent music school — from discovery workshops to a working automation architecture.
Mapped enrolment, scheduling, and parent-communication workflows; designed the automation architecture and phase plan; delivered an interactive client decision portal presenting the solution.
Autonomous Coding Pipeline
An agent system that plans, specifies, builds, and reviews software — from task graph to merged pull request.
Specification generation with acceptance criteria, test-driven implementation, independent multi-axis AI code review as a merge gate, and evidence artifacts for every claim of completion. Built as orchestration and hook engineering on Claude Code with a dependency-tracked task DAG.
Voice-First AI Coaching App
Android · productionAndroid app for assisted voice journaling with AI coaching — a hand-built conversational agent framework, written before adopting industry frameworks.
Custom agent architecture with memory, context management, and tool orchestration; ten coaching conversation strategies with dynamic selection; voice-first core loop chaining listening, reasoning, and speech.
Automated Deep Research Agent
Runs multi-source investigations end to end, verifying every claim into a cited report.
Coordinates multiple research models across providers, evaluates source quality, and synthesizes findings with full citation trails.
Hierarchical Goal Manager
Android · daily useGamified task manager organizing work as a three-level goal tree with focus timers and time banking.
MVVM architecture across 51 Kotlin source files; Room database with 13 schema migrations; six screens under Material Design 3.
Technical Skills
| Languages | Python · Kotlin · TypeScript · SQL |
|---|---|
| Backend & Web | FastAPI · React · Next.js · Docker · Git |
| Android | Android SDK · Jetpack Compose · Room · Material Design 3 |
| Voice AI | Pipecat real-time pipelines · WebRTC · Deepgram · ElevenLabs · Google Speech/TTS |
| Agent Infra | MCP server development · Claude Code orchestration (hooks, subagents, skills) · LangGraph · n8n |
Experience
Founder & AI Engineer — AR Labs
- Deliver AI automation for small-business clients end to end: discovery workshops, process mapping, solution architecture, phased build, and a client decision portal for each engagement.
- Design and ship production systems across three platforms — Android apps in daily use, a live real-time voice agent, and an autonomous coding pipeline with enforced review gates.
- Build in public: technical writing on AI engineering practice, open tooling, and documented architecture for every major system.
Before founding AR Labs: a research foundation in cognitive science and neural-network modelling, followed by a deliberate three-year retraining into professional software engineering — Python and machine learning fundamentals, LLM architecture, Android development, then full-stack web and voice AI systems.
Education & Research
BSc Cognitive Science — First Class Honours
Neural networks, connectionist models, cognitive architecture. Research internship at Carnegie Mellon University with Prof. J.L. McClelland.
McClelland, J.L., & Thompson, R. (2007). Using connectionist models to simulate how children reason about causality. Developmental Science, 10(3), 353–365. DOI
Continuous engineering study
Fast.ai Practical Deep Learning · Andrej Karpathy's Neural Networks: Zero to Hero · ActiveLoop LLM training & fine-tuning · Google Android codelabs · daily hands-on practice across the full stack.
Working Together
I take on contract and consulting engagements — AI automation for operations, voice AI systems, and agent infrastructure — remote, for clients in Europe, Asia, and worldwide.