[SYS.LOG//001]
Generative AI engineering, automation, and creative systems.
CAPABILITIES
Storyboard, image, caption and reference pipelines, orchestrated end to end by Claude Code agents. A brief-to-storyboard CLI turns direction into a shot list; a rate-limited, budget-tracked image service turns it into finished frames.
LLMs, diffusion models and machine learning, run through ComfyUI, HuggingFace and Gradio — vision analysis and generation.
Flux, Kling, Veo and LoRA models, chained through ComfyUI, turn a brief into commercials, UGC ads and stylised video — lip-synced where the shot calls for it. An auto-editor lays word-highlighted captions onto vertical video.
SELECTED WORK
Case study
Case study
METHOD
I start with the workflow as it runs today: what goes in, what has to come out, and which steps are still done by hand. Data cleaning and analysis come before any model is chosen.
Then the system on paper: which model does which job, and what each stage hands to the next. GenFlappy's README sets out the AI model it uses, and why.
Each stage is built as its own system, then connected to the rest. Adcelerate is made this way: SceneBoard, ImageEngine, MoodBoarder and AutoEditor each live in their own repository.
Last, it ships: a service other code can call, or an app anyone can open. GenFlappy runs on Vercel with serverless functions, playable in the browser.
PROFILE
I'm Devanshu Rana, a Generative AI Engineer and a graduate of IIT Roorkee — with a focus on pushing what the technology can do.
I bring a strong foundation in AI, with hands-on experience across today's technologies.
My goal is to build solutions that create real impact.
devanshurana4492@gmail.com
