I am Bishal Paul. I am an AI automation engineer based in the UK, Head of AI at Absolute Intelligence UK, and the founder of Erudience, the small studio behind the client work on this site. Alongside that I run four live products of my own: push8, Extinde, Automation Workflows, and ExcelErrorFinder. Every product I ship on my own name informs how I build for clients, and vice versa.
The category is noisy. Every agency now sells AI automation and most of them have never operated a workflow past the demo stage. What I offer is the opposite. I sit inside the failure modes: FTP feeds that arrive late, webhooks that retry twice, LLM calls that time out under load, phone calls that go to voicemail at exactly the wrong moment. Those are the things a real production system has to handle, and they are what most of my work is actually about.
If you are searching for an AI automation engineer in the UK because you have a specific problem in mind, a stalled internal build, or a founder led AI feature that needs to survive real users, this page will tell you honestly what I do, how I work, and whether we are a fit.
- Design self hosted n8n systems on DigitalOcean and Docker Compose
- Build voice AI agents on ElevenLabs, Twilio, and Vonage
- Ship applied LLM features into real SaaS products with Claude and OpenAI
- Wire AI ticket triage into Freshdesk, Intercom, and Zendesk
- Build LinkedIn and email automation with human in the loop review
- Operate systems in production after they launch, not only build them
- Advise founders on build versus buy for AI features
- Write SEO and marketing systems for AI products
The obvious reason is timezone and language, and those matter more than most founders admit. But the practical reason is that I have already operated the specific stacks UK teams keep landing on: self hosted n8n on cheap European infrastructure, Supabase as the data backbone, Claude for reasoning because of Anthropic's UK friendly stance, ElevenLabs and Twilio for voice because they work well with UK numbering.
The second reason is scope discipline. I run a small studio on purpose. That means I say no to engagements that would fail, and I keep the engagements I take on tightly scoped so they actually finish. Most of my clients hire me a second time for a different system, which is the honest measure that the first one worked.
Live production work covers telecom (voice AI for outbound collections with in call payment capture), retail and e commerce (voice ordering into Shopify, inbound routing to seven departments), energy (multilingual subsidy qualification), local services (weekend and overflow call coverage), financial services and M&A (deal sourcing pipelines with human approval gates), and agency operations (multi brand content pipelines across LinkedIn, Facebook, and Instagram).
Every one of those started as a problem the client was tired of hearing about. The pattern under each is portable: the voice AI for a telecom is the same shape as the voice AI for a solicitor's office, at a fraction of the volume. Half of my job on a new engagement is recognising which pattern already fits.
I turn down chatbot builds that are really an internal knowledge base problem. I turn down full company platform rebuilds under the AI banner. I turn down projects where the client cannot name a single measurable outcome. And I turn down agencies looking for a cheap white label engineer. Erudience is a studio, not a subcontractor.
- Self hosted n8n automation platform
Docker Compose stack on DigitalOcean or Hetzner, Supabase for state, encrypted backups, version controlled workflows, and structured error routing. The client owns the runtime and the data.
- Voice AI agent for inbound or outbound calls
ElevenLabs plus Twilio, with voicemail detection, DTMF handoff for payments or confirmations, multilingual variants where needed, and structured post call records into Supabase.
- Applied LLM feature inside a SaaS product
Claude or OpenAI behind a prompt store, with quotas, versioning, evaluation, and a real pricing page and onboarding flow. Not a demo, a product surface end users pay for.
- AI support triage and internal ops tooling
Freshdesk or Intercom, an LLM reasoning layer, and a Supabase store of product knowledge. Every AI action logged and every escalation to a human traceable to the prompt version.
- Lead generation and follow up automation
LinkedIn and email outreach with genuine reply handling, personalisation, stop conditions, and lead scoring. Human handoff on ready to close conversations.
- 01Discovery call
A thirty minute conversation on the actual problem, the current state, and what a good outcome looks like. Free.
- 02Paid scoping
A short paid scoping engagement, typically one to two weeks, that produces a clear architecture, a fixed build price, and a timeline you can commit to.
- 03Build
I build the system end to end, with weekly demos into a shared environment so you see the workflow taking shape rather than a big reveal at the end.
- 04Launch and hand off
The system goes into production behind your infrastructure, documented so your team can operate it. I stay on for a defined stabilisation window.
- 05Optional retained operation
Many clients keep me on part time to operate the system, add capability, and respond to failure modes as they emerge. This is optional and month to month.
What does an AI automation engineer actually do?+
An AI automation engineer designs, builds, and operates systems that use AI models plus workflow tooling to remove repetitive work from a business. In my case that means self hosted n8n on DigitalOcean, Supabase as the data layer, Claude and OpenAI for reasoning, and voice tooling like ElevenLabs and Twilio when the workflow involves phone calls. The role is different from a data scientist or a general software engineer because most of the value sits in the integration and operational layer, not the model itself.
Are you based in the UK, and do you work with clients outside the UK?+
Yes to both. I am UK based, and Erudience, the studio I run, is a UK entity. Most of my client work is with UK teams, but a meaningful share is US and European. Remote engagements are the default; on site is available for UK clients when the build genuinely benefits from it.
How is this different from hiring an agency?+
Agencies typically resell tools and rely on junior implementers. Every engagement I take on is delivered by me end to end, or by a small trusted network I bring in for specific specialisms. You always speak to the person actually building the system.
What is a typical engagement size?+
The most common shapes are a fixed scope discovery, a defined build project measured in weeks not months, or a retained operator role on a system already in production. I do not do open ended time and materials work.
How do you decide what to automate and what to leave alone?+
Discovery starts with the process, not the tools. If a step is repetitive, rule based, high volume, and has a clear success signal, it is a strong automation candidate. If it needs judgement, involves ambiguous inputs, or hides a broken process, the answer is often to fix the process first. Saying no to an automation is part of the job.