DWG NO. AH-001SCALE 1:1LOCATION LIVERPOOL, UKDRAWN BY A.H.

$ whoami

Alan Hasbiro_

> AI automation and web systems for small businesses, engineered like a build — diagrammed first, shipped after.

> I direct coding agents — Claude Code, Gemini — then test and verify everything they produce before it ships. MSc Digital Marketing. Ran the marketing and operations for two family businesses in Liverpool before this.

./hire-me ./run-demo --live

available for 1–2 new projects

status.log● all systems operational

09:14:02 PASS fact-checker gate — Q2 sector revenue growth = 12%, sourced

09:14:03 PASS content-engine mini — groundedness check clean

09:14:05 INFO 9 automation & AI systems shipped

09:14:06 INFO 3 languages — English, Arabic, Kurdish

email subscribers captured via automation

website traffic from one campaign

automation & AI systems shipped

languages — English, Arabic, Kurdish

FIG. 1

Live demos

not screenshots — run the actual logic, in your browser

Fact-Checked Publisher

A deterministic gate that blocks hallucinated figures before they publish. Type a claim below, or load a sample — it's checked against the same 5-fact bank the real demo uses, running the exact algorithm client-side.

Load a sample or type your own claim above, then click "check claims."

what's in the source fact bank

Q2 sector revenue growth = 12%
Forward P/E ratio for the sector = 18.4
Central bank benchmark rate held at = 5.25%
Unemployment rate reported at = 4.1%
10-year treasury yield closed at = 4.32%

run it yourself, full version →

Content Engine mini

Recording → transcript → persistent per-brand knowledge → grounded content draft → intelligent nurture placement → human approval. Below is real, captured output from the actual program — same recording, one run clean, one with a fabricated figure slipped in.

Content Engine mini — Northfield Property Group (fictional demo brand), recording 001 (good) Retrieved 2/2 knowledge item(s) as relevant context: - case-study-refurb - landlord-deposit-rules Groundedness gate: PASS Today I want to talk about a refurbishment we just finished on a three-bed terrace. PASS We budgeted eighteen thousand pounds for the refurb and finished under budget. PASS The tenant moved in within two weeks of completion. PASS Landlords keep asking about deposit protection rules — that's something we handle for every tenancy. Nurture-slot decision: FILL_EMPTY_SLOT slot 2 (case-study) is empty and matches this content's theme ✅ All claims grounded. Awaiting human approval — nothing publishes on its own.

Content Engine mini — Northfield Property Group (fictional demo brand), recording 001 (bad) Retrieved 2/2 knowledge item(s) as relevant context: - case-study-refurb - landlord-deposit-rules Groundedness gate: PASS Today I want to talk about a refurbishment we just finished on a three-bed terrace. PASS We budgeted eighteen thousand pounds for the refurb and finished under budget. PASS The tenant moved in within two weeks of completion. BLOCK This refurbishment achieved a 40% return within six months of completion. ↳ figure not found anywhere in the transcript or brand knowledge base ⛔ 1 claim blocked. Human approval step is skipped — a blocked draft never reaches nurture placement.

run it yourself, full version →

FIG. 2

How I build

worth being upfront about — it's the actual skill on display here

AI coding agents write most of the code.

I direct the build — requirements, architecture calls, what's non-negotiable, when to reject vs. retry — and I test and iterate on the output. The value I bring is knowing what "correct" looks like and writing the rules that keep an agent honest, not typing every line by hand.

Deterministic decoupling.

LLMs handle language and orchestration; money, safety, and compliance-critical decisions run in plain code. An LLM parses and drafts — it never decides.

Hard programmatic gates, not prompt-tweaking.

Reject malformed or unsupported AI output with real checks — regex, type, source-matching — never trust a model's formatting or a tool's exit code by default.

FIG. 3

Projects

personal / portfolio builds, most at working-prototype stage — click one open for the real decision behind it

01KNOWLEDGEBASE +

Multimodal platform that turns long-form video into viral-ready 9:16 shorts: staged ASR → scene detection → audio-excitement tagging → LLM virality scoring → fusion scoring → captions → B-roll → thumbnails.

Python · FastAPI · Celery · FFmpeg · Whisper

The decision: never pass raw video/audio through the job queue — only <1KB storage keys. Media bytes in a message broker is how these pipelines quietly fall over under load.

02ClipEngine +

Long-form video → platform-ready short-form clips for faceless YouTube channels: transcribe, cut, animate captions, judge quality.

Python · FFmpeg · faster-whisper

The decision: the quality-judge step chains Gemini first with a raw-fallback path, so a failed or malformed AI judgment still returns a usable result instead of blocking the render.

03EspressoE +

Turns a cafe's own phone footage into Reels/TikToks for the business itself: ingest → analyze → cluster into shooting sessions → render → caption → judge → publish.

Python · FFmpeg

The decision: split out of ClipEngine into its own repo and database once the two stopped overlapping — one clips long-form YouTube video, the other clips a single business's own footage. Different job, kept separate on purpose.

04Cafe Mail +

Email marketing + B2B outreach app for a real small business, built to UK PECR/GDPR rules. Foundations (auth, queue, worker) are in; the sending engine is next.

Next.js · TypeScript · PostgreSQL · Docker

The decision: compliance and suppression-list logic is being built before the sending engine, not after — so it's not possible to send a single email that skips consent rules, even by accident, once sending goes live.

05INSTA MESSANGER +

Outreach engine for Instagram-first beauty businesses: prospects, generates a personalised demo page per lead, tracks through a CRM.

Next.js · Convex · Playwright

The decision: the cloud sandbox couldn't launch a headless browser, so scraping moved to a self-hosted runner calling back over plain HTTP — isolating the fragile part instead of forcing it to fit.

06FBA Sourcing Engine +

Finds products sold cheaper at UK retailers than Amazon UK.

Python · AI agents

The decision: the LLM researches and extracts; every profit number runs through deterministic code. A wrong number here is a real financial mistake, not a typo.

07Career-Ops +

This system — a CLI-agnostic AI job-search and freelance pipeline: offer evaluation, CV tailoring, proposal drafting, application tracking.

Next.js · Node.js

The decision: an early batch run fired one agent per item with no limit and hit rate limits hard. Fixed with a proper queue capped at 3 concurrent runs — a lesson learned from a real failure, not avoided in advance.

08Aminata Bar & Bistro +

Mobile-first ordering, booking and loyalty PWA with a swappable mock data layer.

Next.js · Tailwind · React

The decision: a fully clickable demo, including mock checkout, before any backend spend — so the idea can be sold before it's paid for.

09Compliance Engine +

Deterministic allergen computation for food labelling.

Cloudflare Workers · D1 · Hono

The decision: an LLM never decides whether something contains an allergen. If an ingredient isn't positively known, the system refuses to produce a label rather than guess.

FIG. 4

What I offer

fixed packages for small businesses. custom/hourly for anything bigger

package	price
AI Email & Marketing Automation Welcome / nurture / win-back sequences, AI-assisted copy, list setup, basic segmentation.	£150–450	enquire →
AI Short-Form Video Content Scripted, voiced and captioned short-form video, delivered ready to post.	£120–400 per batch of 10	enquire →
Website + AI Assistant Up to 5 pages with an embedded AI assistant that answers customer FAQs and takes enquiries.	£450–900	enquire →
Automation audit & build Process audit + written implementation plan, then n8n/custom automation build.	from £350 then £30–35/hr	enquire →

FIG. 5

About

MSc Digital Marketing (Liverpool John Moores University). Currently a customer service agent at Acorn Insurance, building AI automation projects alongside it. Previously ran the digital marketing and day-to-day operations for two family businesses in Liverpool — a dessert store (automated email + video marketing, 200+ subscribers captured, +30% traffic, +15% sales) and a convenience store (+15% revenue, team of 6, 100% retention).

Native English, Arabic and Kurdish, with 6+ years professional interpreting experience — genuinely useful for bilingual content and outreach work, which is rare on most freelance platforms.