RESEARCH AND DEVELOPMENT
Applied research in AI engineering and cloud platforms.
Nexaaris Labs is our internal research and development function: structured, time-boxed experiments with emerging technology, so that client recommendations are grounded in measured, hands-on evidence.
Why we run a lab
Emerging technologies move faster than their documentation. Before we recommend a pattern to a client, we build it, benchmark it, try to break it and document the results. The lab is where that happens — on our own infrastructure, never on client systems or client data.
Areas we are exploring
Our research agenda follows the questions enterprise clients ask most often.
- Agentic AI: multi-agent orchestration with Microsoft Agent Framework and LangGraph, tool permissioning and Model Context Protocol (MCP) servers
- Evaluation: automated evaluation harnesses for groundedness, task success and tool-call accuracy; LLM-as-judge calibration
- Retrieval quality: chunking strategies, hybrid search, re-ranking and their effect on answer accuracy
- Private inference: vLLM and llama.cpp throughput, quantisation trade-offs (GGUF, AWQ, GPTQ) and GPU sizing for open-weight models
- Platform engineering: reusable delivery pipelines, workload identity federation, policy as code and SLSA-aligned build provenance
- Cost-aware architecture: serverless and consumption hosting for small and spiky workloads
How we work in the lab
Every experiment starts with a hypothesis, a time box and explicit success criteria — quality, latency, cost or operability. Results are recorded in a structured write-up with the configuration needed to reproduce them. Many experiments end with a recommendation not to proceed; that is a valuable result.
From research to client work
Patterns that prove themselves become reusable accelerators — reference architectures, infrastructure modules, evaluation datasets and pipeline templates — which shorten delivery time and reduce risk on client engagements. Broadly useful findings are published in Insights.
What the lab is not
The lab is not a product line and does not process client data. Anything that graduates from an experiment into client work goes through the same architecture, security and delivery discipline as the rest of our engagements.
Contact: contact@nexaarislabs.com