HomeStacksProjectsAboutContactResume
Back to Index
Hugging Face APIFastAPIReactPydantic

CreatorGPT

The Problem

Content creators spend hours brainstorming titles, drafting captions, researching hashtags, and conceptualizing thumbnails. Standard AI workflows forced them to jump between multiple tools and manually stitch assets together.

The Hard Constraint

The system had to deliver a complete, cohesive package—text copy plus 3 unique thumbnail image concepts—in under 10 seconds to feel instant, bypassing standard sequential API bottlenecks.

Architecture

[User Brief] │ ▼ [Text LLM (Forced JSON Object Mode)] │ ├──► copy └──► image_prompts: [p1, p2, p3] │ ▼ asyncio.gather() ├──► [HF Image Model (p1)] ├──► [HF Image Model (p2)] └──► [HF Image Model (p3)] │ ▼ [Package Assembled & Schema Validated] │ ▼ [UI Rendered in < 10s]

Results & Caveats

The Outcome

Designed an asynchronous two-stage pipeline using Python's asyncio.gather to execute three text-to-image API requests entirely in parallel. Cuts content preparation time by 90%.

Honest Caveats

Relying on free Hugging Face inference endpoints leads to 402/502 errors during peak traffic. Additionally, open-source image models still struggle with generating perfectly legible text directly on thumbnails.