Inspiration

My idea was to create a knowledge pipeline for myself.

What it does

The app researches practical workflows or tips for predefined programs and creates videos that end up on adjustable social media channels after approval via Telegram.

How we built it

BrandPulse, a TypeScript monorepo built with Next.js, Fastify, Prisma, SQLite, Remotion, FFmpeg, and Tauri, employs a local worker to coordinate various tasks like research, generation, asset capture, voice synthesis, rendering, approval, and publishing. Runtime reasoning is now divided among Hermes specialist profiles. GPT-5.6 handles topic research, editorial decisions, script development, scene planning, and evidence review using the OpenAI Codex provider. GPT Image 2 generates source visuals, while Qwen3-TTS produces the cloned brand voice. Remotion and FFmpeg assemble the final video. Codex served as the primary engineering environment, enabling me to trace failures across the API, worker, renderer, Telegram workflow, and publishing integrations. It also helped implement the scene director, browser and application capture, typography fitting, TTS quality checks, OAuth validation, channel-safety checks, and tests around these paths.

Challenges we ran into

The connection from all Social Media Platforms, Video quality, repetitive topics

Accomplishments that we're proud of

I used this app to manage three knowledge YouTube channels and my own social media presence for Zolahrsounds.

What we learned

Each stage needs different context and checks. Research needs sources, writing needs the brand voice, scene planning needs timing and assets, and publishing needs account identity and approval.

What's next for BrandPulse

implement tiktok and linkedin

Built With

Share this project:

Updates