Prologue

On shipping conversational AI
to production.

A note on what fbmhdl does, how we do it, and what it costs. Last updated May 2026.

I. Practice

A note on how we work.

Most conversational AI today is bolted-on no-code chained to a vendor's platform. It looks impressive in a sales deck and falls apart the first time a customer asks something the demo didn't cover. We do the opposite.

We write software in your stack, with your data, your guardrails, and your engineers in the room. The repository is yours from day one. If we walked away tomorrow, the assistant would keep running.

We work in fixed two-to-twelve week engagements. We don't sell licences, recurring "AI seats," or anything that looks like a SaaS tier. You pay us to build something. Then you own it.

An assistant is the easy part. The wiring is the work.
II. What we build

Three kinds of project.

i.

Customer-facing assistants

Support, account questions, order tracking, the long tail of "talk to a human." Deployed to web, WhatsApp Business, Messenger, Telegram, voice, or your own mobile apps.

ii.

Internal copilots

Domain-trained agents for ops, sales, or back-office teams. Connected to your CRM, ticketing, knowledge base, and internal APIs. Audit trail and role-based access by default.

iii.

Integration & rollout

The unglamorous middle layer: telephony, payments, identity, evals, logging. We do this on its own for teams that have built the assistant but can't get it into production.

III. Live work

Two assistants in production.

Both demos are running on the same patterns we'd use for a paying client. They handle real interactions, not curated happy-path examples.

Logistics · courier service Live

ABC Parcel

Customer support for a courier. Handles tracking, address changes, delivery rescheduling, customs questions, and clean handoff when the request needs a human.

abc.fbmhdl.com →
Retail banking · onboarding & support Live

Banking assistant

Account-aware assistant for retail banking. Product explanations, onboarding walk-throughs, scoped tool calls, full PII scrubbing and audit log for every interaction.

demo.fbmhdl.com →
IV. Process

Three phases. No retainer trap.

1.

Discovery — week 1, fixed fee

We map the workflow you want to automate, the data the assistant needs, and what "good" looks like. Output is a written scope and a working prototype.

2.

Build — weeks 2 to 8

We build against your real channels and integrations. Weekly reviews, staging environment from day one, an eval suite so changes don't quietly break.

3.

Hand-over

Code in your repo. Docs in your wiki. A short handover with your engineers. We're available afterwards on retainer or by the hour, but you don't need us to be.

V. Practical details

The boring questions.

§

Who owns the code?

You do. It lives in your repository under your licence from day one.

§

Which model do you use?

Whatever fits. Frontier APIs from OpenAI or Anthropic for most projects; open-weights on your own infrastructure when the data demands it.

§

Can you work with regulated data?

Yes. We sign DPAs, work inside your VPC, and use private model endpoints. We've built for banking before.

§

What does it cost?

Discovery week is £3,500, fixed. A full build is £18,000 to £65,000 depending on scope. We quote on the actual project.

VI. Correspondence

Write to us.

A few lines about what you're building is enough to start. We reply within one working day. If we're not the right fit, we'll say so and suggest who is.

@

Email

hi@fbmhdl.com

Phone / WhatsApp

+44 7783 990485

Office

182-184 High Street North, London E6 2JA