What we build
From a single automation to a multi-agent AI pipeline
Custom builds, off-the-shelf integrations, AI where it earns its place, deterministic automation where it does the job better. The work scales from a single Zapier step to a custom agentic team with persistent memory and proper state management.
Below are common patterns. They are illustrative, not exhaustive. The right answer for you'll sit somewhere on that spectrum.
Demos
A few worked examples
Yours won't look like any of these. That is the point.
The tool is never the point. The fit is.
AI Systems
When the input is unstructured, AI earns its place
Inbound messages, free-form documents, customer questions, judgement calls in the middle of a workflow. This is the work where rules-based automation gets brittle. We use LLMs, RAG pipelines, and agentic patterns where the cost of being slightly wrong is acceptable, and a deterministic alternative would be heavier to build than the problem deserves.
AI agents and multi-agent orchestration
- An inbox triage agent that reads incoming emails, classifies them, routes to the right person, and drafts replies for the most common patterns.
- A multi-agent pipeline that takes a customer inquiry, researches the account, drafts a tailored quote, and flags anything unusual for human review.
- An ops agent that monitors a queue or system status and acts on a defined set of recoverable issues before paging anyone.
RAG chatbots and context engineering
- An internal chatbot that answers staff questions using your own documentation, policies, and past tickets, rather than a generic ChatGPT giving generic answers.
- A customer-facing assistant grounded in your product catalogue, returns policy, and live inventory, with hand-off to a human when it doesn't know the answer.
- Context-engineered prompts and retrieval pipelines designed so the model has exactly the information it needs, and nothing it doesn't.
Document and email classification, extraction, and routing
- Inbound emails auto-categorised, tagged, and routed before they hit the inbox.
- Invoices and receipts auto-extracted and posted into your accounting system, with exceptions flagged for review.
- PDF or contract field extraction into structured data the rest of your stack can use.
Voice agents
- An after-hours call-answering agent that takes messages, answers FAQs, and books callbacks.
- A voice front-end on an existing booking, ordering, or support flow, with structured handoff back to your team.
LLM-powered internal tools
- A drafting assistant for proposals, quotes, or email replies, fed with your tone of voice and recent examples.
- A summariser that turns long meetings, threads, or documents into structured action lists posted into Slack or Notion.
- A classifier that sits between your form and your CRM, enriching each record before it lands.
Agentic teams with persistent memory and state orchestration
The deep end. Engineering-grade work for problems a single agent or prompt chain can't hold.
- A team of specialised agents (researcher, drafter, reviewer, executor) coordinating through shared memory and a state graph, handling work that spans hours or days and persists context across the cycle.
- Custom orchestration built on LangGraph or similar agent frameworks where agent flow, error handling, retries, and human-in-the-loop checkpoints are first-class concerns, not afterthoughts.
- Memory management designed to fit the use case (short-term, long-term, episodic, semantic, retrieval-augmented) so the system actually remembers what matters and forgets what doesn't.
- Goal-driven agentic operations: take a high-level objective, plan the steps, execute across your tools, learn from outcomes, report back.
Scoped per problem. Larger than a single Scoped Build, often a Foundations Reset or a defined multi-stage engagement.
Automation & Integration
When the rules are stable, automation is the right answer
Most of the time the problem isn't "we need AI", it's "the same information has to live in three places and they don't agree". Deterministic automation is faster to build, cheaper to run, and easier to trust than dropping an LLM into the middle of an invoice flow. We pick the platform after the workflow is clear, never before.
n8n workflows
- Self-hosted n8n where data sovereignty matters: customer data, financial flows, internal records that shouldn't leave your environment.
- A new customer in HubSpot triggers contract generation, signature capture, onboarding sequence in Slack and Notion, and CRM update, all in one flow.
- Mid-complexity automations where Zapier hits its limits but a fully custom build is overkill.
Zapier and Make integrations
- Lower-volume, well-supported flows where speed-to-build matters more than fine control.
- Quick wins in the early days of a foundations rebuild while the deeper work is being designed.
API integrations across the SME stack
- Connecting two tools that don't have an off-the-shelf integration: a niche CRM to a bespoke quoting system, an industry-specific platform to your accounting software.
- Replacing a fragile chain of Zaps with something owned, documented, and version-controlled.
- Direct integrations with Microsoft 365, Google Workspace, Xero, HubSpot, Salesforce, Notion, Slack, Stripe, and most things with a public API.
Custom code automations
- Node, Python, or TypeScript when the workflow needs logic, error-handling, or scale beyond what a no-code tool can hold.
- Self-contained scripts and services that live in your own environment and do exactly one job, well.
Webhook and event-driven flows
- One system tells another what just happened, automatically, in near-real-time. Less polling, less duplication, less drift between sources of truth.
Custom Builds
When nothing off-the-shelf quite fits
Most SMEs don't need a bespoke product. Some do. Where an existing tool would force the business to bend around it, a small focused build is often cheaper over five years than the workarounds that accumulate around a near-fit SaaS.
Internal tools and lightweight web apps
- A small internal app that lets the team trigger jobs, see system status, or update records without going through a developer.
- A spreadsheet outgrown into a real tool, with proper validation, history, and access control.
Reporting, analytics, and dashboards
- A real-time view of operational KPIs pulled from multiple systems and refreshed automatically.
- Replacing the weekly spreadsheet that someone hand-builds every Monday morning.
Scheduled jobs and batch processes
- Nightly data exports, recurring report generation, automated reconciliations.
- Long-running tasks that don't need a person watching them, set up so they fail loudly when something goes wrong.
Not sure what bucket you fit in?
None of the above quite fits?
Most of the work we end up doing started as "I'm not sure if this is even a thing you do." If you can describe the friction in plain English, we can usually tell you within a day whether it's a fit and what it would cost.
The principle, again
Tool follows from problem, never the other way around
We'll tell you when AI is the right answer, and when a single Zapier step does the job better. We'll tell you when off-the-shelf fits, and when custom code is worth the effort. We'll tell you when the problem is automation, and when the real problem is that the underlying process needs cleaning up first.
The deliverable is a system that runs, integrated into your existing ecosystem, documented, owned by you. Whatever underlying tools we used to build it.