SYDNEY, AUSTRALIA

AI engineering. Built fast. By people who use it daily.

Small Sydney shop. We design, build, and ship production AI systems — agents, automations, full-stack apps. Most of what other teams quote in months, we ship in weeks.

Three things, done well.

AI & LLM

AI & LLM Systems

RAG pipelines, multi-agent orchestration, voice agents, custom tooling on top of Claude/OpenAI/local models. We build the boring infrastructure around the model so the model can do its job in production.

Full-Stack

Full-Stack Web

Next.js + FastAPI + Supabase. Auth, payments, dashboards, integrations. Same stack we use to ship our own product, so we know where the cliffs are.

Automation

Workflow Automation

End-to-end: ingest → enrich → score → route → act. Replace the spreadsheet, the Zapier maze, the part-time VA, all at once. Our own pipeline runs 24/7 — happy to show you.

Three weeks, not three months.

01

Day 1–3

We get on a call. You explain the problem. We tell you whether it's worth building, what we'd charge, and how long. No deck.

02

Week 1

Working prototype. Deployed. You can poke it. We adjust.

03

Week 2–3

Production hardening. Tests. Monitoring. Hand-off doc. Done.

Our main tool is Claude Code. That's not a gimmick — it's why we ship in weeks instead of quarters. We treat the AI as a junior engineer that doesn't sleep, and we review its work properly.

We built Agency OS for ourselves first. It runs every day.

15+
AI Agents
5
Distribution Channels
10
Tier Enrichment

If you want a deeper walkthrough, we'll demo it live. No NDA needed — we'd rather you see it work than read about it.

Selected work.

Recent projects we've delivered.

01

Agentic AI Copilot for Financial Operations

Financial services firm · AI-powered portfolio management interface

Built an agent layer on a read-only DB with tool-use API. A planner agent decomposes natural language queries, a SQL agent generates statements, a validator runs EXPLAIN before execution, and an executor returns structured results. Tool registry handles action verbs — rebalance, alert, export. Shipped in 4 weeks.

Python/FastAPI Claude API Postgres pgvector Docker
Try Live Demo →
User Query Planner Agent SQL Generator Validator EXPLAIN Results + Actions input decompose generate safe? execute
02

RAG Search Over Internal Docs

Professional services firm · Internal knowledge search platform

Built a hybrid search system over 50k+ internal documents. Ingestion pipeline chunks and embeds into a hybrid index combining BM25 keyword scoring with dense vector retrieval. Reranker resolves conflicts. Answer synthesis returns cited spans with source highlighting. Cut internal lookup time by 80%. Delivered in 6 weeks.

Python Postgres + pgvector OpenAI Embeddings FastAPI Next.js
Try Live Demo →
Docs source Chunker + Embedder ingest Hybrid Index BM25 | Vector dense retrieval Reranker score Answer + Citations verified
03

AI-Powered Outbound Sales Pipeline

Australian B2B agency · Multi-channel outbound automation platform

Built a complete outbound pipeline — discovery feeds an 8-tier enrichment waterfall, CIS scoring picks the channel (email, voice, LinkedIn, SMS), and reply detection feeds back into scoring. The model keeps learning from real signal. Processes thousands of prospects daily across 5 channels. Shipped in 3 weeks, running in production since.

Python/FastAPI Next.js Supabase Redis Resend Vapi Telnyx Prefect
View Dashboard →
Discovery GMB / SERP Enrichment 8 tiers waterfall CIS Score 0–100 Channel Router select Email Voice LinkedIn SMS reply signal loops back
04

Full-Stack SaaS for Vertical AI

Healthcare technology startup · Patient management SaaS platform

Delivered a multi-tenant SaaS platform — Next.js frontend over FastAPI backend, Supabase for auth with row-level security for per-tenant isolation. Stripe for billing. AI features integrated as inline tools — summarise, score, suggest. Auth + dashboards + first AI feature live in 3 weeks. Full v1 in 6.

Next.js FastAPI Supabase Stripe Claude/OpenAI Vercel + Railway
Next.js Frontend Vercel / CDN FastAPI Backend Railway Supabase Auth (RLS) Postgres Storage AI Models Claude / OpenAI Stripe billing / webhooks
05

Custom Workflow Automation Engine

Data analytics company · Custom workflow orchestration engine

Built a Prefect-based workflow engine with a custom step library replacing their n8n/Zapier stack. Visual pipeline builder lets non-engineers wire steps without YAML. Built-in cost tracking, retry logic, and dead-letter queue. Observability via OpenTelemetry. Engine + first 5 step types delivered in 3 weeks, visual builder in 2 more.

Python Prefect FastAPI Next.js React Flow Postgres OpenTelemetry
Try Live Demo →
Visual Builder React Flow drag + drop Prefect Engine retry / DLQ cost tracking orchestrate Step Library API Call Transform AI Step Notify Observability — OpenTelemetry traces + metrics across all steps

Sydney-based. Small team. AI-native.

Sydney-based. Small team. AI-native — we use the same tools we sell.

We take on a handful of projects at a time. Each one gets our full attention — no account managers, no handoffs to juniors. You work directly with the people writing the code.

ABN on file. SPAM Act compliant. Australian English throughout (we don't "leverage" things).

The hard questions.

We're a small team. The AI tools are how a small team punches above its weight. You're not paying for headcount — you're paying for what gets shipped.
Same answer as any engineering shop: tests, monitoring, rollback, on-call. The model writes code. Humans review it. Same review bar as a good engineering team.
Small team is the point. The systems you're hiring us to build — we already built for ourselves. Agency OS runs 15+ agents in production daily. We're not learning on you — we're applying what already works.
Depends on scope. Hourly is $150–220 AUD. Fixed-price is more honest for projects with clear scope. Ask.
Same week, usually. Email below.

Let's talk.

Replies same day.

david.stephens@keiracom.com