SN

Flagship AI Product

YouTube Autopilot

I built a deployed AI workflow automation platform that turns a video niche into a planned, generated, edited, optimized, and upload-ready YouTube video.

Live product AI Products

Live

deployed product

End-to-end

AI workflow

Async

job orchestration

Secure

credential handling

Context

What existed before my work

This is the product and technical situation I was responding to, not just a list of tools.

Video creation is not one task. It is a chain of decisions: topic research, scripting, visual planning, voiceover, editing, metadata, publishing, and cost control.

I wanted this project to prove more than prompt writing. I wanted it to show that I can build a product-grade AI system with users, jobs, settings, credentials, status, and deployment concerns.

System Map

How the work connects

I want each project page to show the system, not only describe it.

Research

Script

Visuals

TTS

Edit

SEO

Upload

My Ownership

Exactly what I owned

These are grouped by responsibility so the case study reads like real product ownership, not scattered bullet points.

Product and UX

  • Designed the app as a control center for AI video jobs rather than a one-command script.
  • Built job creation, dashboard, settings, analytics, and live progress surfaces.
  • Made cost and pipeline progress visible so the user is not waiting on a black box.

AI workflow

  • Modeled the pipeline as specialized responsibilities: planning, script refinement, visuals, TTS, editing, SEO, and upload.
  • Used a CEO-style orchestration pattern to turn a vague niche into a structured production plan.
  • Separated planning, generation, media assembly, and publishing concerns.

Backend and deployment shape

  • Built around FastAPI, job records, Redis/ARQ background tasks, Postgres models, and artifact storage.
  • Handled user credentials through encrypted storage and OAuth flows.
  • Designed per-job output isolation so generated files and statuses can be tracked cleanly.

Key Modules

The parts of the system that carry the story

Each module is a concrete proof point: product surface, backend, integration, workflow, or leadership responsibility.

AI agent pipeline

The pipeline moves from research and planning into script, visuals, voiceover, editing, SEO, and upload. The important part is not only generation; it is coordination and observability.

CrewAIPlanningSora/TTS flowSEO

SaaS dashboard

The frontend gives users a product surface for sign-in, API key setup, YouTube connection, job launch, progress, history, analytics, and settings.

Next.jsTypeScriptDashboardSettings

Async worker system

Long-running generation work is handled outside normal web requests, with records and state updates that make the pipeline trackable and safer to operate.

FastAPIRedis/ARQJobsArtifacts

Credential and publishing flow

The product stores user API keys and YouTube tokens carefully, then uses those credentials to run generation and upload workflows without exposing secrets in the UI.

EncryptionOAuthYouTube APISecurity

Hard Problems

Constraints, tradeoffs, and fixes

This is where the case study becomes credible: what was difficult, and how I responded.

AI media work is slow and unpredictable.

I separated the request/response app from the worker pipeline and exposed live job state so the product stays usable while generation runs.

A user needs trust before letting automation publish content.

I treated dashboard visibility, cost tracking, settings, OAuth, and status feedback as core product features rather than extras.

A demo can look impressive while still being fragile.

I shaped the project as a SaaS-style system with auth, settings, jobs, persistence, artifacts, and deployment instead of stopping at a local script.

Leadership

What I carried beyond code

This is my current proof that I can combine AI engineering with the product and full-stack instincts I built before my MS program.

I used this project to practice the kind of AI product work I want to do professionally: not only model calls, but complete workflows with user-facing control.

Lessons

What this proves about how I think

AI products need product surfaces, not just prompts.

Long-running AI workflows need observability, state, and failure-aware architecture.

My full-stack background is what lets me turn AI ideas into deployed systems.

Stack

Technologies and integrations

Next.jsTypeScriptFastAPIPythonCrewAIPostgresRedis/ARQCloudflare R2YouTube APIOpenAI

Next

This is the level of ownership I want to keep building from.

My strongest work happens when I can move across product, engineering, integrations, and AI workflows with real responsibility.

Back to projects