AI Product Builder & Forward Deployed Engineer

Find where the work gets stuck, build AI-native products and workflows that fix it, and deploy them inside organisations with approvals and guardrails built in.

Have a process to automate, or an AI system that needs proper governance and observability? Glad to talk.

Start a conversation Play tour

Directed four products, each with a stage stated plainly. Nothing released without approval.

Act 01 of 04

Isolate

Find the one place the work gets stuck.

A process fails in one place first

Everything looks slow from outside. Usually one handoff, approval or manual task holds up the rest. Finding it is most of the work.

Start with the stuck point

Where does work wait? Which step is always manual? Who chases whom? Writing the answer as a testable problem before anything is built.

Act 02 of 04

Trace

Follow the fix from objective to approval, one stage at a time.

Stage 1

Objective and context

Defining the objective and supplying the context. A build that starts without both is guesswork.

Stage 2

Break it into stages

Work runs through APIs, integrations, Model Context Protocol (MCP) servers and retrieval-augmented generation (RAG) for source-based tasks. Agentic steps cover research, data processing, drafting and checks.

Stage 3

Review apart from building

Code review, evaluation and testing sit apart from implementation; CI runs before anything is published.

SpecifiedLogging and observability record what ran, what changed and what was approved.

Stage 4

A person signs

Human gateNothing is published, sent or paid for without approval, and every claim must be backed by evidence.

How a build is governed

The method, in full

Treating each build as a governed process: development, observability, and forward deployed engineering once live. Defining the objective, supplying context, breaking the task into stages and reviewing every result before it moves on.

Development is agentic. Work runs through APIs, integrations, MCP servers and RAG for source-based tasks. Agentic workflows cover research, data processing, drafting and checks; some have browser and computer access, each with a defined role, coordinated by an orchestrating step. Comparing results, correcting errors and asking for evidence before accepting a claim.

Review stays human-in-the-loop. Code review, evaluation and testing sit apart from implementation; CI runs before anything is published.

Governance is data handling, approvals and guardrails. Sensitive material stays on local or self-hosted models, not outside services. Logging and observability record what ran, what changed and what was approved, so a live product can be fixed in place. The same loop ships PitchReach, TextBackCaller, AMLtracker, the Global Investor Guide and the Drew & Crew.

Act 03 of 04

Release

Put it to work where the work is.

Ship it where the work happens

A workflow that only works in a demo is not a fix. It has to run inside the organisation, on its real data and under its real rules.

Then stay with it

A live system is connected in place, then watched and corrected after it ships. That is the forward-deployed part of the job.

Guardrails built in

Data handling, approvals and guardrails are set before a workflow runs. A change is checked and logged, and it does not go out without approval.

What can be done for you

From the stuck point to something running

  • AI-native products

    Directed products from objective through to a reviewed result. Shipped this way: PitchReach, AMLtracker and the Global Investor Guide.

  • Business automation

    Workflows for the routine steps of a business.

    • Missed-call and enquiry follow-up
    • Lead capture and CRM sync
    • Appointment booking and reminders
    • Invoice, quote and payment chasing
    • Document and data extraction
    • Scheduled reporting
  • Scalable autonomous workflows

    Work is split into stages, with a person setting the objective and checking the result.

  • Orchestration frameworks

    Several steps are coordinated so each handoff stays explicit.

  • Forward-deployed integrations

    A live system is connected in place, then watched and corrected after it ships.

  • Governance and guardrails

    Data handling, approvals and guardrails are set before a workflow runs. A change is checked and logged, and it does not go out without approval.

Case study

TextBackCaller

Live

AI-assisted automation, end to end: from a missed call to a text back, with a person deciding at the points that matter.

ShippedA missed-call text-back service. Live, launched 19 October 2026.

ShippedFor UK small businesses.

ShippedCustomers keep their existing business number.

The automation

Once, when a customer signs up

  1. 1Customer signs up and pays by cardStripe Checkout
  2. 2A signed payment webhook starts provisioningCloudflare Worker
  3. 3A UK mobile number is bought for the customerTwilio API
  4. 4Setup text and setup email go outTwilio SMS, Resend
  5. !If setup fails or a trial is refused, the owner is alerted and a person decidesHuman gate

Every missed call

  1. 1The phone network diverts the missed call to the customer's numberConditional forwarding
  2. 2A signed webhook arrives; the call is rejected as busy at onceWorker, Twilio voice
  3. 3Guards are checked, then one text goes to the caller from the business's own numberTwilio SMS, D1

ShippedOne text per caller per 24 hours, UK mobile numbers only, and the owner's own calls are skipped.

ShippedSignup, payment, number provisioning and setup messages run automatically, with no manual step between a paid signup and a working service.

Tools called

Services the system calls, and what for
ServiceCalled for
Stripe Checkout and BillingTaking the subscription; signed payment webhooks back
Twilio Voice and SMSBuying numbers, receiving missed-call webhooks, sending the text
Cloudflare WorkersRunning the logic, with a scheduled job each hour
Cloudflare D1Signups, customers, usage, opt-outs
ResendSetup and alert emails

ShippedEach service is called through its API: Twilio for voice and SMS, Stripe for checkout and billing, Resend for email, D1 for storage.

ShippedBuilt on Cloudflare Workers, D1, Twilio Voice and SMS, Stripe Checkout and Billing, and Resend.

Written in

  • JavaScript
  • SQL
  • HTML
  • CSS

ShippedWritten in JavaScript on the Cloudflare Workers runtime, with SQL in D1 for data, and HTML, CSS and browser JavaScript for the pages.

Built as AI-assisted automation

ShippedBuilt as AI-assisted automation: AI workers implement each staged release and call tools through APIs and connectors. Reviewed the evidence and approved.

Where a person decides

Human gateEvery release needs approval. Builds are staged, tested, backed up and given a rollback point first.

Human gateNew features ship switched off. They are switched on in a separate, approved change, and a kill switch turns each one off again.

Human gateWhen automatic setup fails or a trial is refused, the owner is alerted so a person decides what happens next.

ShippedRelease controls: signed webhooks, duplicate-send prevention, opt-outs, usage caps and automated tests.

Also shipped

The rest of the work

Directed each of these under Drew & Crew Ltd: set the objective, supplied the context and reviewed every result.

PitchReach

Beta

ShippedA football opportunity map of England for parents, guardians and coaches, in controlled beta.

ShippedJourney planning by car and bus is open to everyone.

SpecifiedUses age, postcode, transport mode and travel time to assess proximity and practical access.

SpecifiedArchitecture designed around Next.js, TypeScript, MapLibre, Python, FastAPI, PostgreSQL, PostGIS, GeoPandas and routing services.

SpecifiedControlled beta access and safeguards against bulk extraction were defined.

AMLtracker

Live

ShippedA free calendar of upcoming UK Part-66 exam sittings, for people studying for the aircraft maintenance licence. Dates come from the UK CAA and Part-147 centres, and no account is needed.

ShippedEvery date on the live calendar links to its source.

Global Investor Guide

Expanding

ShippedA published, evidence-led global investing guide, being expanded.

ShippedIncludes a tax-residency ranking of 25 countries.

LimitNot financial advice. Sources, assumptions and limitations are stated.

Drew & Crew Ltd

Live

ShippedThe Drew & Crew Ltd company site, live, listing 9 ventures.

In development

A problem being worked on

Concept

AssumptionA focus tool that helps students actually put in the study hours they plan (concept).

Act 04 of 04

Contact

Have a process to automate, or an AI system that needs proper governance and observability? Glad to talk.