Keep the essentials working.
My background includes identity and access, Google Workspace, endpoint management, and enterprise infrastructure. At loveholidays, that work supported more than 1,350 internal and external users.
AI Automation | System Engineer
From a messy workflow
to a system
people can use.
I’m Arafaat Shaikh, an AI automation specialist and system engineer based in London. I connect AI, APIs, and cloud infrastructure to turn complex operational work into dependable tools.
My work spans customer workflows, event-driven pipelines, document intelligence, and the interfaces that make them usable.
The work, from end to end
UNDERSTAND
A real operational problem
↗
↓
ENGINEER
APIs + AI + infrastructure
✳
↓
OPERATE
Review. Observe. Recover.
↻
Workflow & integration
Events, APIs, customer context
Applied AI
Extraction, synthesis, validation
Cloud engineering
Queues, state, deployment
Operational reliability
Logs, retries, reconciliation
01 / My background
Systems first.
Automation grew from there.
I started in IT operations and systems administration. Working with the same provisioning tasks and disconnected tools pushed me towards workflow development and, later, applied AI.
Connect the repetitive work.
I connected HiBob, Google Workspace, and Freshservice to automate employee lifecycle tasks, then built API and webhook integrations across operational tools.
Build on that foundation.
I now work on voice and chat processing, structured summaries, and event-driven cloud pipelines. Systems experience shapes how I handle access, data integrity, failures, and recovery.
02 / Selected work
The problem.
The decisions. The system.
Public builds and anonymised work. Internal data and proprietary implementation details are omitted.
ANONYMISED CASE STUDY / CUSTOMER OPERATIONS
Bring voice and chat context into the customer workflow.
Customer context arrives through CRM events, chat transcripts, and voice attachments. The engineering challenge is coordinating that work and returning a validated summary to the tools people already use.
My contribution
Designed an event-driven ingestion and synthesis pipeline, including asynchronous processing, validated LLM summarisation, reconciliation, and audit visibility.
GCP · Python · Gemini · AssemblyAI · BigQuery
Receive & decouple
CRM webhook → gateway → Pub/Sub
Queue & track
Cloud Tasks + Firestore lifecycle state
Transcribe & synthesise
Voice transcription + validated LLM output
Return & observe
CRM write-back + BigQuery audit trail
Reconcile
Scheduled checks for stale events
The decision
Separate receiving from processing.
Decoupling webhook ingestion from transcription and synthesis lets the work move through an asynchronous pipeline.
The reliability work
Account for unfinished work.
Lifecycle state, audit logging, and scheduled reconciliation make stale events visible and support recovery.
The resulting capability
Validated context in existing tools.
The pipeline returns synthesised customer context to enterprise tools, with an audit trail for investigation.
Raw media to a shareable video.
A Streamlit workflow that combines vehicle media and a script into a vertical video, handling narration, clip timing, aspect ratio, and export.
Inspect the source ↗Long documents to reviewable context.
A document analysis workflow combining session-scoped ingestion, structured summaries, and source context in a review interface.
What makes this an engineering problem?
Keeping sessions separate, coordinating multiple documents, and retaining source context all matter alongside the model’s response.
03 / Explore the work
Try the flow.
Ask about the decisions.
Run the live NodeFlow pipeline and ask the Gemini-powered portfolio assistant about my work.
See what happens between input and output.
Follow validation, parsing, extraction, schema checking, and output without digging through a log stream.
LIVE SANDBOX · Input is processed by the NodeFlow service. Use sample text, not private information.
Open the original live NodeFlow sandbox ↗Ask about my work.
Start with a question, or choose a topic below.
I can explain the projects, engineering responsibilities, and reliability approach described in this portfolio.
GEMINI ASSISTANT · Questions are sent to the AI service. Curated answers are available if the service is unavailable.
04 / How I work
The work continues
after the first successful run.
Start with the handoff.
Understand what someone is trying to do, where the context lives, and what makes the task difficult before choosing a model.
Connect with care.
Make event contracts, validation, state, and review part of the workflow so each component has a clear responsibility.
Build for investigation.
Use logs, retries, alerts, and reconciliation to understand what happened and find work that needs attention.
05 / Let’s connect
Have a workflow
that needs engineering?
Let’s talk about applied AI, customer operations, and the systems that connect them.