Let's Talk

Selected Work

AI systems, automation, and software we've built.

Explore the business problem, what we delivered, and how each system works. Filter by industry to find projects relevant to your team.

18 projects across all industries

01

Healthcare & Social Care

Regulated clinical and welfare systems, where compliance is an architectural constraint rather than a checklist.

Healthcare · HIPAA · Platform Engineering

HIPAA-Capable Behavioral Health Platform

The Challenge & What We Built

A US behavioral health provider had outgrown a legacy system that could serve one organisation and could not be safely extended to more. We rebuilt the clinical platform on Google Cloud with the compliance boundary drawn explicitly: BAA-covered services, isolated environments, encryption and audit logging throughout, and least-privilege production access. Cutover was rehearsed on a data copy before a live record was touched, and phase two adds true multi-tenancy, a Super Admin control plane, and self-service onboarding.

What Was Delivered

HIPAA-capable architecture

Multi-clinic expansion scoped as a next phase

Governed scope and change control

Migration rehearsed on a data copy

Built With

  • GCP
  • Firestore
  • Cloud Run
  • Identity Platform
  • Pub/Sub
  • Claude Code
  • React

Social Services · Azure · Twilio

Foster Care Placement Outreach System

The Challenge & What We Built

A child welfare provider was placing children through manual, call-by-call outreach, a process where speed materially affects outcomes and no record survives the phone call. Now an anonymised child profile is submitted, the system returns a ranked shortlist of foster parents, and a specialist assigns priority through a deliberate human gate before any message leaves. SMS broadcasts run in priority order, replies are captured through a secured webhook, and every contact is timestamped and escalated when a window closes unanswered.

What Was Delivered

Faster placement cycle

Human judgement preserved

Every contact evidenced

Client-owned data residency

Built With

  • Azure App Service
  • Azure SQL
  • Azure Functions
  • Twilio
  • Microsoft Entra ID
02

Real Estate & Property

Document-heavy operations where analyst hours and lease turnaround decide the margin.

Real Estate · Document AI

Real Estate Document & CRM Intelligence

The Challenge & What We Built

A real estate investment firm was losing expensive analyst hours to document handling and manual CRM entry. We scoped a phased automation programme sequenced so value arrives in week one rather than month three: an immediate flyer-generation win, then CRM auto-population through a property data API, then an AI review pipeline for Letters of Intent and Purchase & Sale Agreements that extracts and checks terms against the firm's own positions, and automated triage of inbound listing email.

What Was Delivered

Analyst hours back on deals

LOI / PSA document AI

CRM auto-population

Value in week one

Built With

  • LLM Agents
  • Locate Alpha API
  • Document Extraction
  • Gmail API
  • n8n

Document Automation · PropTech

Automated Lease Generation Pipeline

The Challenge & What We Built

Approved rental applications arrive by schedule or webhook, data is fetched from Buildium and validated, leases are generated as HTML, converted to PDF, filed to Drive, logged in Sheets, and emailed to the leasing team.

What Was Delivered

Fully automated

Zero manual entry

End-to-end error handling

Built With

  • n8n
  • Buildium API
  • JavaScript
  • Google Drive
  • HTML-to-PDF

Sales Intelligence · Multi-Agent AI

AI Sales Call Auditing & Coaching Engine

The Challenge & What We Built

Managers at a student housing operator were reviewing sales calls by hand, which in practice meant most calls were never reviewed and the few that were got judged inconsistently. Recordings are collected continuously, transcribed, and filtered before two parallel scoring tracks run: an agent chain assessing opening, requirement discovery and objection handling, alongside an engagement agent rating rapport with verbatim evidence. An aggregation stage resolves both into a single comparable score, written back per caller into the CRM.

What Was Delivered

Every call held to one standard

Coaching from evidence

Comparable caller benchmarks

CRM write-back

Built With

  • OpenAI GPT-4o
  • gpt-4o-transcribe
  • Python
  • Google Drive API
  • CRM API
03

Retail & E-Commerce

Brand, catalogue and customer lifecycle systems built to run at volume without diluting the voice.

Autonomous Content · SEO · E-Commerce

Autonomous SEO Content Engine

The Challenge & What We Built

A London fine jewellery house needed editorial volume without diluting a carefully held brand voice. The pipeline runs the full editorial process: caching the live catalogue, planning topic clusters against commercial priorities, programmatic SEO research, blueprinting, then long-form writing assembled into publication-ready HTML for Shopify. A dual-model strategy routes planning to a lighter model and writing to a stronger one, articles generate in parallel, and an automated British English pass holds the register on every piece that ships.

What Was Delivered

Volume without headcount

Dual-model cost strategy

Parallel generation

Brand voice enforced

Built With

  • Claude Code
  • Claude Sonnet + Haiku
  • Python
  • Shopify API

Email Marketing · Agentic AI

AI Email Marketing Agent for Klaviyo

The Challenge & What We Built

The same jewellery brand was writing campaign emails by hand and staging them through Gmail drafts, disconnected from the platform that actually sends. The agent turns a short campaign brief into a finished, on-brand email: production-ready HTML with inline styling, a dynamically composed hero with layered imagery and typographic overlay, and safe fallbacks for the mail clients that break modern layout, pushed straight into Klaviyo as a reviewable draft. The marketing lead's role shifts from writing to approving.

What Was Delivered

Brief to draft in minutes

Writing time reclaimed

Direct Klaviyo draft push

Human approval retained

Built With

  • Anthropic Claude
  • Python
  • Klaviyo REST API
  • HTML Email Engineering

Order Lifecycle · CRM Automation

End-to-End Order & User Lifecycle Automation

The Challenge & What We Built

Order webhooks trigger HubSpot updates, SendGrid transactional mail, lifetime metric updates, and branching logic for first-time versus repeat orders. Parallel onboarding flows create customer and chef profiles and run 1-day feedback and 14-day reactivation sequences.

What Was Delivered

Full lifecycle coverage

First vs. repeat order logic

Automated re-engagement

Built With

  • n8n
  • HubSpot API
  • SendGrid
  • Webhooks
04

SaaS & Connected Products

Multi-tenant platforms, live device infrastructure and the security work that keeps them defensible.

Publishing Platform · Technical Partner

Technical Partner: ManuScripts Publishing App

The Challenge & What We Built

We are the technical partner behind ManuScripts, a publishing application in active development. The engagement covers product engineering and platform ownership under a standing retainer: feature design and delivery, release management, and the architectural judgement that keeps a growing product coherent rather than accreted. Working as an embedded partner rather than a ticket queue means the roadmap conversation and the implementation sit with the same people.

What Was Delivered

Embedded technical partner

Product engineering ownership

Release and maintenance ownership

Active development

Built With

  • Ongoing retainer
  • Feature development
  • Release management

Connected Devices · Security & Infrastructure

Technical Partner: Connected GPS Tracking Platform

The Challenge & What We Built

We are the technical partner behind a Dubai company's connected tracking product: mobile app, Laravel API, CMS, and a persistent ReactPHP TCP service holding thousands of live device connections. A full codebase audit surfaced a hardcoded master-password backdoor, a committed signing key, and plaintext card storage carrying real PCI-DSS liability. Rather than hand over a report, we turned the findings and the client's commercial priorities into one prioritised remediation programme and took ownership of the work. Now a standing monthly engineering retainer.

What Was Delivered

Payment data exposure closed

Remediation owned, not just reported

AWS target-state estate designed

Ongoing engineering retainer

Built With

  • Laravel
  • ReactPHP TCP
  • MySQL
  • Redis
  • React Native
  • AWS ECS Fargate
  • CloudFront

Chatbot Development · AI Agents

AI-Powered Customer Support System

The Challenge & What We Built

We built a customer-support agent connected to the client's CRM and knowledge base to answer inbound questions using their business information.

What Was Delivered

Automated support responses

Knowledge-base integration

CRM integration

Built With

  • OpenAI GPT-4
  • LangChain
  • Python
  • FastAPI
  • React
  • PostgreSQL
05

Finance & Professional Services

Research, sourcing and qualification pipelines for firms whose product is information and reach.

Content Research · Multi-Agent AI

Reddit Content Intelligence System

The Challenge & What We Built

A pipeline monitoring four Reddit communities, classifying discussions with Claude into four financial categories, and generating YouTube-ready content ideas with titles, angles, and outlines into Notion. Runs daily at 8 AM.

What Was Delivered

4

Specialist Agents

4

Communities

Daily

Cadence

Built With

  • Claude
  • Reddit API
  • Notion API
  • n8n
  • Python

Lead Generation · Productised System

Multi-Industry Lead Generation Engine

The Challenge & What We Built

A Dubai consultancy wanted a lead engine it could point at a new industry without commissioning a new build each time. Sourcing runs across multiple independent scrapers with cross-referencing and deduplication, so no single provider becomes a point of failure or a pricing hostage. Contact resolution escalates through a waterfall of enrichment providers, stopping at the first verified result. Where no database holds the answer, AI research agents read company sites and public sources to complete the fields vendors simply do not carry.

What Was Delivered

New vertical by configuration

Coverage up, cost per record down

AI research agents

Client runs it independently

Built With

  • Python
  • Anthropic Claude
  • Apollo
  • SalesQL
  • Lusha
  • Bright Data

Sales Automation · CRM Integration

Sales Lead Qualification & Scheduling Bot

The Challenge & What We Built

We built an agent that qualifies leads from multiple sources, enriches contact details, and books meetings with qualified prospects.

What Was Delivered

Lead qualification

Meeting scheduling

CRM integration

Built With

  • Python
  • LangGraph
  • Anthropic Claude
  • Zapier
  • HubSpot API
  • Salesforce API
06

Media, Recruiting & Field Operations

Publishing throughput, campaign operations and on-the-ground evaluation, run by system rather than by headcount.

Recruiting · Meta Ads · GDPR

Automated Meta Recruiting Campaign System

The Challenge & What We Built

A German recruiting media agency was running client campaigns through manual ad operations that could not scale beyond the hours of the people doing them. We delivered the feasibility assessment, architecture and statement of work for a Python engine that builds, launches and manages Meta recruiting campaigns end to end, with Monday.com as the operator interface so nobody learns a new tool. GDPR and EU data residency were architectural constraints from the first design decision, not a compliance retrofit.

What Was Delivered

Capacity beyond operator hours

EU data residency by design

No new tool to learn

Phased build underway

Built With

  • Python
  • Meta Marketing API
  • Monday.com API
  • Hostinger VPS
  • Google Drive API

Social Media Automation · AI Scheduling

Multi-Platform Video Caption & Scheduling Engine

The Challenge & What We Built

New Drive uploads are detected, validated, batched, and passed through Gemini for optimised multi-platform captions. Clients are matched to profile groups, publishing slots computed, posts scheduled, and results aggregated into an emailed HTML report.

What Was Delivered

Multi-platform scheduling

AI-generated captions

Structured error alerts

Built With

  • n8n
  • Gemini AI
  • Lomavis API
  • GCS
  • Gmail API

Meeting Intelligence · AI Evaluation

Automated Meeting Intelligence & Scoring System

The Challenge & What We Built

Zoom recordings arrive via HMAC-validated webhooks, AWS Lambda processes transcripts, and multi-agent pipelines score meetings across engagement, opening, closure, action items, and effectiveness.

What Was Delivered

5 scoring dimensions

Automated transcript analysis

Meeting evaluation workflow

Built With

  • n8n
  • Zoom Webhooks
  • AWS Lambda
  • OpenAI GPT-4o
  • Google Sheets API

Computer Vision · Field Operations

Video-Based Cleanliness Evaluation System

The Challenge & What We Built

A Telegram bot captures on-ground cleaning videos, pairs them with reference benchmarks, and analyses both with Gemini to assess debris coverage by zone, logging weighted scores and video links to Sheets for audit.

What Was Delivered

AI zone-level scoring

Adaptive scoring logic

Centralised audit trail

Built With

  • n8n
  • Telegram Bot API
  • Google Gemini
  • Google Sheets API

Have a similar challenge?

Tell us what's slowing your team down or what you want to build. We'll help you work out the next step.

Get a Free AuditRequest a Discovery Call
Start a ConversationWork · ScaleWorks AI