We ship AI software that earns its keep.
For SaaS founders and operators who need AI in production — not in a slide deck. We’ve built 37 products ourselves — email triage, AI tutors, content engines, CRM tools — so your project gets proven infrastructure from day one. First working demo in 1–2 weeks.
Client work
Not our product — their business. Paying engagements where we’re the outside build or SEO/GEO team.
Edwards Kirby Lawyers
Full website rebuild for a Sydney insolvency and business law firm, plus a secure client portal — JWT + MFA, encrypted document storage, SEO/GEO built in from day one.
About to go live Read the story → Client · SEO/GEOShuttersmith
The first client on our productised SEO + GEO service — technical audit, structured data, and generative-engine optimisation so the business is found by Google and by AI assistants.
354 real visits, majority via Google · led to a second business Read the story → Client · Home improvementNew Shutter Business
A second business, launched off the back of Shuttersmith’s SEO/GEO results — trade-price plantation shutters with a self-install option and a full measure-and-quote build.
In staging, launching soon Read the story →This is for you if…
You have a workflow that still runs on people, not software — and you want to change that without a six-month engineering project.
You know where AI could help but don’t have the team to build the eval pipeline, prompt ops, and infrastructure that make it reliable.
You know what an LLM is. You know the business case. You need a team that’s actually shipped it in production — not just demo’d it.
Why Dainty Trading
Most AI agencies spend eight weeks writing a deck. We spend it writing code.
Proof of work, not pitch decks
We have 37 products — 17 live and paying — you can click on right now. Every project page shows the real stack, the build timeline, and what broke. That’s the portfolio. Judge us on it.
Speed that’s earned, not promised
First working demo in 1–2 weeks. First paying customer in 4–8 weeks. We move fast because we’ve already built the boring parts — billing, auth, LLM routing, observability — and reuse them across every project.
Proven plumbing, not glued-together demos
We’ve already built the layer most AI projects trip over — routing, billing, observability, evals. Every client gets it on day one, because we run it for our own products first. Stack: Claude · LiteLLM · n8n · FastAPI · Postgres
We align on P&L, not feature lists
AI that doesn’t reduce headcount, cut support tickets, or generate revenue isn’t AI — it’s a line item. We build to measurable outcomes and tell you what success looks like before we start.
How an engagement works
Hard deadlines at every stage. No discovery phase that lasts longer than the build.
Brief received
Fill in the contact form. We reply with our read on the project and any clarifying questions — same business day.
Scope agreed
If it’s a fit, we send a one-page scope for the first sprint: fixed price, hard end date, and exactly what you’ll have at the end of it.
First demo live
A working slice of your product, deployed somewhere you can use it. Not a mockup, not a Figma — running software.
A few of the 37
Five live products, from years of daily use to fresh launches. See the full portfolio →
BrightPath
Personalised K-12 online school with an AI tutor named Pax, gamified rewards, and progress tracking that actually nudges kids.
AI tutor in production · K–12 students on-platform daily Read the story → AI · ProductivityEmail Triage
Claude-powered email categorizer that scores priority, drafts replies, and runs across multiple Gmail and Outlook accounts.
200+ emails processed daily · Runs across Gmail & Outlook Read the story → AI · ContentGhost Writer
End-to-end content engine: detects trending topics, writes 1,500-word articles, runs 6-layer quality checks, and auto-publishes to LinkedIn and Medium.
3 articles/week published · No human writer in the loop Read the story → Finance · AutomationSolar Loan Reconciler
Built for a financial company. Collects Mon–Fri solar loan transaction CSVs, matches them against the weekly Loan Payout Report, flags any gaps, and emails a clean HTML report to the loans team every Friday.
Zero missed loans since go-live · Fully automated every Friday Read the story → AI · BotsNudgle
Telegram reminder bot with natural-language scheduling. Quiet hours, vacation mode, recurring tasks — all parsed by Claude.
Read the story →AI Development Services
We don’t do AI demos. We ship AI features that show up in your P&L. Four things, done well.
AI product builds
Full-stack delivery. We take an idea — yours or ours — and ship a paying product. Backend, frontend, mobile, billing, observability, and the LLM glue that makes it real.
Automation retrofits
Already have a SaaS or internal tool? We bolt in AI where it earns its keep — email triage, support deflection, content generation, structured extraction — without rewriting the world.
AI infrastructure
LLM gateways, prompt versioning, evals, queue workers, billing plumbing, observability. The unsexy layer that decides whether your AI feature stays up at 3am.
SEO & GEO optimization
Get found by Google and by AI answer engines. We audit and fix meta tags, schema markup, llms.txt, and AI-crawler access — shipped directly to your codebase, not a PDF. Free audit, Setup from $2,900 AUD — not the $5k–$15k-plus retainer agencies typically quote for the same audit-and-fix scope.
How we work
No retainers, no long SoWs to sign. Three steps from brief to paying customers.
Brief → scope in 48 hours
Tell us the workflow or product in two paragraphs. We come back with a realistic scope: what ships in sprint one, what the stack looks like, and what it costs. No proposal theatre.
Sprint one → demo in 1–2 weeks
We don’t start with architecture reviews. We ship the smallest thing that proves the idea works, put it in front of a real user, and let you decide whether to continue. One working feature. Deployed. Clickable.
Iterate to a product → weeks 3–8
Each sprint adds one feature that earns its keep. Daily written updates. Weekly live demo. You own the roadmap; we own the execution. By week eight, you have paying customers or we’ve told you why.
Frequently asked questions
Short answers to what people ask before booking a call.
What is Dainty Trading?
Dainty Trading is an AI automation studio. We design, build, and operate production AI software — 37 products covering email, content, education, knowledge management, marketing automation, scheduling, billing, and SaaS infrastructure. Some are shipped and live; others are in final testing or active development. Each project page is honest about which.
What kind of AI projects do you take on?
LLM-powered features (Claude, Gemini, OpenRouter), agentic workflows, content automation pipelines, AI-driven personalisation, and structured-extraction systems. We also do the boring infrastructure that keeps them up — gateways, queues, evals, billing.
What stack do you build on?
Python (FastAPI, Django) and Node.js (Fastify, Express) on the backend; Next.js, React, and React Native on the front; PostgreSQL and Redis for state; n8n for orchestration; Stripe for billing; LiteLLM for routing across model providers; Docker everywhere.
Do you only build with Claude?
No. We default to Anthropic Claude because it punches above its weight on long-context and structured tasks, but we route through OpenRouter and LiteLLM so projects can pick Gemini, OpenAI, or open-weights when the economics or latency demand it.
How fast can you ship?
First demo in 1–2 weeks for most builds. A first paying customer in 4–8 weeks. We move fast because we’ve already built the boring parts (billing, auth, LLM routing, observability) and reuse them across every project.
Where are you based?
Australia. We work async with clients globally and are happy to operate in your timezone.
How do I start?
Email us via the contact page with two paragraphs: what you want built and why it matters. We’ll reply within one business day with a scoped first sprint.
From the blog
How we think about building AI in production.
How long does it take to build a production AI agent?
The demo takes a week. The gap to reliable production is evaluation pipelines, fallback handling, and edge cases — here’s the honest timeline.
Read → EngineeringRules Still Win: When Not to Use an LLM
LLMs aren’t a silver bullet. Our decision tree for when deterministic rules outperform AI models — and when to use both.
Read → CostToken Cost Optimization: Where the Savings Actually Are
The highest-ROI strategies for reducing LLM costs in production: prompt caching, model routing, context trimming, and output constraints.
Read →What users say
“We engaged Andrew to have a look at our website,he explained what he needed to do and fixed everything and we our now getting lots more jobs through from.the internet traffic He also built a new website for a new branch of the business The work is fantastic and he absolutely nalied it Its about time we spent some money and got the return you hope for After several companies I have found the one I will be sticking with”
“SATs week and we have been using BrightPath for about 2 weeks now. Ethan just learnt long division, in the space of a day and actually enjoyed the lessons. He asked me why didn’t school teach me this way — the step-by-step guides at the start help so much. Thanks for the help BrightPath team, keep up the good work.”