Debug/Fix AI Video Generation Pipeline – FFmpeg, Node/Python

I am building a video automation tool that converts scripts into fully edited videos using AI-generated visuals (images + short video clips). I am looking for an experienced backend developer with strong knowledge of: Video processing pipelines Asynchronous job systems AI API integrations Important: This project already has an existing codebase. Your first task is to review, debug, stabilize, and improve it not rebuild from scratch. Current Issues (Must Be Fixed First) The system is currently unstable. Key issues include: Projects intermittently failing to load (dashboard inconsistency) Media uploads failing randomly or not appearing in the timeline B-roll fetched is often irrelevant to scene prompts FFmpeg rendering pipeline failures Uploaded clips appearing as still images in final render. Your First Responsibilities: Debug and stabilize the system Identify root causes (not temporary fixes) Ensure reliability before any new features are added Core System Upgrade (After Stabilization) We are moving away from stock footage APIs toward AI-generated visuals. New Workflow: The system should: Take existing scene prompts (already generated by the tool) Generate visuals automatically using: Image generation models (I will decide which one to use) PixVerse (text-to-video) Scene Handling Logic The backend should intelligently decide: Motion scenes (movement, animals, action, disasters, rare events) → Generate 5-second video clips via PixVerse (looped) Static scenes → Generate images + apply Ken Burns effect User-Controlled Inputs (Frontend → Backend) When a user selects a video length, they must also specify: Number of image scenes Number of video/clip scenes Example: Video length: 15 minutes Image scenes: 60 Video scenes: 30 The backend must: Distribute scenes accordingly Override intelligently when needed (On the backend, the tool should determine which scenes require generated video clips and which require images. Scenes involving movement, animal actions, disasters, rare events, or anything that requires motion should automatically be assigned to PixVerse. Other scenes can use image generation with cinematic motion effects such as ken burns.). Media Placement: Generated visuals must be automatically placed on the timeline Each video clip must be: Max 5 seconds Looped to fill scene duration (cost optimization) User Capabilities Users should still be able to: Upload/replace custom media via the upload button already in the tool Search for alternatives via the built-in search engine already in the tool. Tech Stack (Current): n8n Backend: Node.js / Python Video Processing: FFmpeg Queue System: BullMQ / Redis Storage: S3-compatible (Backblaze / Supabase) AI Integrations: Image generation + PixVerse Deliverables: Stable, production-ready backend Reliable upload and media handling system Fully functional rendering pipeline (no silent failures) Clean, maintainable, documented code. Important Note I am looking for someone who can: Take ownership Think structurally Build for scale and reliability. Only submit a proposal if you are fully comfortable delivering this project within the budget stated on the job post. Do not apply with the intention of renegotiating later. Respect for scope and pricing alignment is very important for this role.