/ The Lab — Research & Thought Leadership

We share
what we learn.

Deep-dives on AI systems, infrastructure, voice agents, and the specific engineering decisions that separate good software from systems that hold up under real load.

Our publishing principles

We ship, then we write.
Everything here comes from production experience. No hypotheticals.
We name our mistakes.
The most useful articles include the three things that nearly broke the project.
We give you the exact decisions.
Not "consider using Redis" — we tell you exactly why we chose it at this scale.

/ Library— 5 articles

AI Systems★ Featured

Why Sub-500ms Voice Latency Changes Everything in Enterprise AI

The difference between 300ms and 1,200ms in a voice AI isn't a technicality — it's the line between a natural conversation and a frustrating machine. We documented the architecture decisions that got Lobby under 500ms, and why most voice agents never will.

May 20268 min

Architecture

The Case for Edge-First Architecture in 2026

Regional latency is the new conversion rate. We analysed production deployments and found a direct, measurable correlation between edge proximity and user engagement. Here's the data and how we architect every new project accordingly.

April 202612 min

Product Engineering

Flutter vs. React Native: A Decision Framework Based on 20+ Apps

After shipping 20+ mobile applications in both, we have a clear internal framework. Flutter wins on performance and design fidelity. React Native wins on web convergence and team reuse. Here is exactly when we choose each — and when we choose neither.

March 20266 min

Operations

Zero-Downtime Migrations: The Exact Playbook We Follow

We have migrated live production platforms with zero customer-facing downtime. Feature flags, traffic shadowing, automated rollback, and the specific order of operations that has never failed us. Including three times it nearly did.

February 202610 min

AI Research

Production-Grade RAG Systems That Actually Work at Scale

RAG systems are trivial in demos and painful in production. After building multiple production RAG pipelines, we have documented what the tutorials omit: chunking strategies that degrade over time, retrieval quality that collapses with vague queries, and the hallucination patterns that slip through most evals.

January 202615 min

/ About Nolgy

An engineering team
that ships real products.

Nolgy was founded on a simple premise: software companies that never build their own products eventually lose their edge. We build for clients, and we build for ourselves.

Lobby began as an internal experiment. The question: how quickly could we build a voice agent that passes for a trained human on a customer service call? Six weeks. It is now a commercial product.

Talk to us about your project →

2022

Founded

Active products

Countries served

Stack layers

We sharewhat we learn.