# Production AI, engineered to hold up.

neumatic is the applied-AI practice of Greg Shingles. I design and ship AI systems — retrieval, agents, and the LLM infrastructure around them — for teams that need them evaluated, observable and maintainable, not just demonstrated. A decade of production .NET and TypeScript sits behind the AI.

Applied AI · full-stack engineering · Australia (AEST), working worldwide.

## A system I run — HeySomm

HeySomm is a live wine platform I designed and run end to end. Scan a label, and a vision model, retrieval over pgvector, and an eight-axis match engine turn it into a recommendation, streamed back over SSE. The part I'd defend hardest isn't the model work, it's the discipline around it: six scan stages run on versioned, A/B-routable pipelines, and nothing ships until it has been replayed through the real inference path against recorded scans. That is, the AI is treated like any other production system. It runs on a Hono API and a native iOS app, with multi-provider inference — Anthropic, OpenAI, Cohere — behind circuit breakers, deployed blue-green on Fly. It holds up.

Visit heysomm.ai — https://heysomm.ai

## What I take on

- **RAG systems that cite their sources.** Retrieval you can audit — lexical or vector, with reranking and typed citation verification — so an answer points at something a person can check.
- **LLM features that ship with an eval gate.** Multi-provider inference behind one abstraction, versioned prompts, and a replay harness that blocks a release until it has been re-run against recorded cases.
- **The application around the model.** I'm a full-stack engineer first — API, data model, queues, auth, deployment — so the model arrives inside a system that already works.

## Selected work

- **HeySomm** (Cofounder, founding engineer · Live) — A live AI wine platform I run as cofounder and its only engineer.
- **Untangled.Legal** (Solo · In development) — A legal-AI retrieval platform I'm building toward a commercial release: lexical search with citation verification, for regulated text.
- **Enterprise systems** (Team) — A decade of production .NET: order orchestration across a dozen storefronts into 3PL and ERP, and personalisation storefronts for major beverage brands.

## About

I'm Greg Shingles. For ten years I built production systems in .NET and TypeScript — order orchestration, print automation, multi-tenant SaaS — and for the last several I've built AI products, HeySomm among them. My view on AI work hasn't changed since I started: if it isn't evaluated, it isn't finished.

## Stack

| Area | Tools |
| --- | --- |
| Languages | .NET · TypeScript · Python |
| AI / LLMs | Anthropic · OpenAI · Cohere |
| Retrieval | Postgres · pgvector · rerank |
| Infrastructure | Fly.io · blue-green · eval gates |

## Start a project

I take on a small number of engagements at a time — a RAG system, an LLM feature, or the full-stack build around one. Tell me what you're trying to ship and where it's stuck.

Available for contract work — remote, based in Australia (AEST), flexible across US/EU/APAC.

Email: greg@neumatic.io

---

> Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). AI crawling, indexing, and training are expressly permitted.