Research & Experimentation

The AR Labs Research Lab

Where we experiment, break things, and discover what's next. The Lab is where ideas become architectures before they become products.

Research Areas

What We're Exploring

Six active research tracks driving the next generation of AI products.

Agentic Systems

Multi-agent pipelines and autonomous AI workflows with full observability.

MCP Integrations

Building and publishing Model Context Protocol servers and clients.

RAG Architectures

Hybrid retrieval pipelines — BM25, dense vectors, and reranking — for production.

Automation Workflows

Intelligent event-driven automation connecting AI decisions to real-world APIs.

AI Experiments

Novel applications of language and vision models tested in controlled sprints.

Developer Research

DX improvements, open-source tooling, and benchmarks shared with the community.

Active Projects

Current Experiments

Projects actively being researched, prototyped, and validated.

🕸️
Research
Agentic Systems

Agent Mesh

A multi-agent orchestration framework where specialised AI agents collaborate, delegate sub-tasks, and execute long-horizon pipelines with full observability and retry logic.

Multi-agentLangGraphOrchestration
Learn More
🔩
Prototype
MCP Integrations

MCP Forge

A toolkit for building, testing, and publishing Model Context Protocol servers. Includes a type-safe SDK, hot-reload dev server, and a hosted registry for sharing MCP tools.

MCPProtocolSDK
Learn More
📚
Active
RAG Architecture

RAG Core

A modular, production-grade RAG pipeline with hybrid BM25 + vector search, cross-encoder reranking, and support for multi-modal document ingestion (PDF, DOCX, images).

RAGEmbeddingsReranking
Learn More
⛓️
Research
Automation Workflows

AutoChain

Event-driven automation engine that bridges AI model decisions with real-world API calls, webhooks, and database writes — with full audit trails and rollback support.

Event-drivenWebhooksAudit
Learn More
👁️
Experiment
AI Experiments

Neural Vision

Computer vision experiments covering real-time object detection, document OCR, receipt parsing, and visual Q&A — packaged as lightweight inference APIs.

VisionOCRInference API
Learn More
📊
Prototype
Developer Research

LM Bench

An open benchmarking suite that evaluates LLM providers on latency, cost-per-token, instruction-following, and reasoning across hundreds of standardised prompts.

BenchmarksLLM EvalOpen Source
Learn More

Open Source Contributions

We open-source selected lab experiments, CLI tools, and SDKs. Follow our GitHub for the latest releases.

View GitHub
AI · INNOVATE · BUILD · SCALE

Ready to Build?

Whether you have a product idea, need an AI system built, or want to explore what's possible — we respond to every message personally.