Selected Systems
Examples of systems Shah has built — ecommerce first, then the rest of the engineering. Each shows the problem, the system, and what it connects; they are examples of implemented work, not verified client outcome studies. A separate section covers reference architectures: illustrative designs shown with synthetic data.
By the numbers
Real outcomes, not vanity metrics
A track record built on Upwork and direct work.
Fixed-price
scoped and agreed before we build
Amazon Marketplace Reconciliation
Amazon FBA / Marketplace Finance
Problem
Amazon FBA sellers reconcile multi-country tax across large SKU catalogs by hand in spreadsheets — slow and error-prone.
The system
A system that ingests Seller Central data, parses every transaction, aggregates across marketplaces, runs the tax and reconciliation logic, and writes structured reporting back out.
How it operates
- 1.Amazon Seller Central data ingestion
- 2.Transaction parsing
- 3.Bulk aggregation across marketplaces
- 4.Tax and reconciliation logic
- 5.Structured reporting output
What it connects
- Amazon Seller Central
- Ingestion
- Parsing
- Aggregation
- Tax / reconciliation
- Reporting
System function
- Marketplace reconciliation, handled end to end
- Seller Central data ingested and parsed
›Under the hood — the workflow as built
Also in the toolkit
Other systems we've built
Voice AI, RAG, content and data systems from earlier work — the same engineering, pointed at other problems.
Faceless Reel Pipeline (AI Cartoon Generator)
Content / AI Video
Problem
Producing consistent faceless cartoon video content at scale is slow, and characters drift between scenes.
The system
An automated reel pipeline that generates scripted faceless videos with consistent characters across scenes.
What it connects
- Claude
- Flux
- Seedance
- ElevenLabs
- AssemblyAI
- Higgsfield Soul
System function
- Keeps characters consistent across scenes
- Automated scripted-video assembly
›Under the hood — the workflow as built
AI Content Production System (V9)
AI Content Platform
Problem
Manual YouTube/social content production is slow — scripting, recording and editing eat a content team's week.
The system
An automated content pipeline that runs from idea to upload, with the scripting, generation and assembly handled end to end.
How it operates
- 1.Google Sheets Trigger for content ideas
- 2.Anthropic Claude for intelligent script writing
- 3.MiniMax Script Node for video generation
- 4.Video Generation with automated assembly
- 5.Google Drive Upload for asset management
What it connects
- Google Sheets
- Anthropic Chat
- MiniMax Video
- Video Assembly
- Google Drive
System function
- Runs from idea to upload
- Automated script generation and assembly
›Under the hood — the workflow as built
HubSpot Voice AI Agent
SaaS Marketing Agency
Problem
High inbound call volume meant missed calls, and the sales team was overwhelmed with unqualified leads after hours.
The system
A voice agent that answers and qualifies after-hours calls and logs each one to HubSpot, so leads are handled instead of missed.
How it operates
- 1.Webhook for incoming call routing
- 2.HubSpot Get Contact for lead lookup
- 3.Time Window Check for business hours
- 4.Make Call with Twilio integration
- 5.Extract Call Details for transcription
- 6.Update HubSpot with call summary
What it connects
- Webhook
- HubSpot
- Time Check
- Twilio
- AI Transcription
- CRM Update
System function
- Answers and logs after-hours calls to HubSpot
›Under the hood — the workflow as built
Multi-Model Executive Assistant
Enterprise Team
Problem
Overwhelming administrative tasks and deep research requirements for team members. Context-switching between tools wastes significant productivity.
The system
Automated the research-to-task workflow, removing manual context-switching.
How it operates
- 1.Slack Trigger for voice/audio commands
- 2.Transcribe Audio for meeting notes
- 3.SerpAPI Web Search for research
- 4.Google Calendar for scheduling
- 5.Fibery Task Execution
- 6.Supabase Storage for knowledge base
What it connects
- Slack
- Audio Transcription
- Web Search
- Calendar
- Task Manager
- Database
System function
- Turns voice commands and research into tasks
›Under the hood — the workflow as built
Jotform-to-WordPress AI Publisher
Content Publisher
Problem
Content publishing bottlenecks between user submissions and live WordPress posts. Editorial team overwhelmed with 100+ daily submissions.
The system
Automated publishing from form submission to live post, with SEO metadata generated along the way.
How it operates
- 1.Jotform Trigger for new submissions
- 2.OpenAI Text Parser for content extraction
- 3.DALL-E Image Gen for featured images
- 4.WordPress REST API for post creation
- 5.SEO metadata auto-generation
What it connects
- Jotform
- OpenAI Parser
- DALL-E
- WordPress
- SEO Optimizer
System function
- Publishes form submissions to WordPress
- Generates featured images and SEO metadata
›Under the hood — the workflow as built
Browser-Use Automation (Retool)
Data Services Inc.
Problem
Repetitive browser-based tasks that lack native APIs — web scraping and data entry done by hand, week after week.
The system
Reliably automated non-API browser workflows that previously ran by hand.
How it operates
- 1.Webhook trigger for automation requests
- 2.Run Agent Request via Retool
- 3.Extract Execution Info from browser
- 4.Get Video From URL for media tasks
- 5.Supabase Bucket Upload for storage
What it connects
- Webhook
- Retool Agent
- Browser Extraction
- Media Processor
- Cloud Storage
System function
- Automates non-API browser workflows
›Under the hood — the workflow as built
Reference Architectures
How the system works
Illustrative architectures using synthetic or sanitized data. Not presented as live client deployments or observed performance results.
Shopify + Amazon Operations Reference Architecture
DurranAI reference architecture
Reference architecture · illustrativeProblem
An ecommerce team on Shopify and Amazon runs ad monitoring, inventory checks and marketplace reconciliation by hand, with a person pulled into every routine review.
The architecture
A DurranAI-designed workflow showing how an ecommerce team can monitor reporting, inventory, advertising and marketplace exceptions while holding consequential decisions for human approval. Illustrated with synthetic data, not a live client account.
What it connects
- n8n
- Claude API
- Postgres
- Shopify API
- Amazon SP-API
- Telegram
What it demonstrates
- Shopify and Amazon inputs
- Scheduled monitoring
- Missing / stale-data detection
- Exception classification
- Daily operational memo
- Approval queue
- Integration-failure alerts
- Human-approval boundaries
›Under the hood — reference workflow (illustrative)
Have a bottleneck like one of these?
Book an ecommerce teardown and we'll map the highest-impact system for your business — fixed scope and price, honest about what it takes.