From raw media to a ready-to-share video.
A Streamlit tool that turns uploaded vehicle images or footage and a script into a vertical video with generated narration. It handles clip timing, aspect ratio, audio, and the final export in one flow.
ENGINEERING AI FOR REAL WORKFLOWS
Ideas are easy.
Useful systems
take engineering.
I’m Arafaat, an Applied AI & Automation Engineer. I connect language models, APIs, data, and cloud infrastructure to make complicated work simpler and more reliable.
FIG 01 / THE WORKFLOW HUMAN-IN-LOOP
01 / INPUTSignals & contextEvents · documents · APIs
02 / INTELLIGENCEReason + routeModels · tools · guardrails
03 / OUTCOMEUseful actionReview · decide · automate
BUILDING AT THE INTERSECTION OF AI, PRODUCT & SYSTEMSSCROLL TO EXPLORE ↓
01 / SELECTED WORK
Built for the messy
middle of real work.
Public code where available. Private work is described at a high level, with client and internal details removed.
A grounded assistant for complex service work.
Designed an assistant workflow that gathers relevant customer context, uses model reasoning to produce a concise summary, and passes the result into an existing human review process.
What made it an engineering problem
Connecting multiple systems safely mattered as much as the prompt: structured tool inputs, clear boundaries around sensitive data, and a review step before any consequential action.
Make long documents easier to act on.
Built a document analysis prototype that turns lengthy case material and conversation history into structured summaries, with attention to source context and separation between sessions.
What made it an engineering problem
The work spans ingestion, model orchestration, access boundaries, and an interface that lets a person inspect and use the result. The goal is faster understanding while keeping judgment with the reviewer.
Keep the workflow working after launch.
Connected events, APIs, and cloud services into operational workflows, then investigated failures and added clearer retry and alerting paths. The focus: making automation understandable when something goes wrong.
What made it an engineering problem
A useful workflow needs dependable handoffs, good failure signals, and a safe way to recover. I work across the application and infrastructure layers to make those pieces fit.
02 / HOW I WORK
The model is one part
of the system.
The best applied AI work starts with a real user task and ends with something people can trust and operate.
01 /
Start with the job
Understand the decision, handoff, or repetitive task before choosing a model or framework.
02 /
Connect the pieces
Make APIs, data, prompts, interfaces, and deployment work as one coherent product.
03 /
Design for reality
Build in review, visibility, and recovery so the system stays useful beyond a demo.
03 / ABOUT
Curious about the
whole stack.
My work sits between product thinking and implementation. I enjoy finding the small, practical way AI can improve a workflow, then building the integrations and infrastructure that make it usable.
I’ve worked across conversational AI, document workflows, automation, cloud services, and media generation. I’m interested in roles where I can own the path from a useful idea to a working system.
APPLIED AIAUTOMATIONPYTHONAPIsGCPPRODUCT ENGINEERING
04 / LET'S CONNECT
Have a hard problem
worth building for?
I’d like to hear about applied AI, automation, and product engineering opportunities.