Senior Video-Pipeline Engineer - Build an AI Short-Video Editing Service

Title: Senior Video-Pipeline Engineer — Build an AI Short-Video Editing Service (Auto Captions, Cuts, Zooms, Music) — Multilingual incl. Hebrew & Arabic (RTL) About us We run a growing SaaS (web + mobile app) that creates and publishes social-media content for business clients. Today our video editing runs on a third-party API (Submagic/OpusClip-class tooling). We want to bring this capability in-house as our own service, integrated behind our existing backend. What we're building A server-side pipeline: it receives a video, edits it automatically, and returns a finished 1080×1920 MP4 via webhook. Capabilities, in priority order: Milestone 1 — Captions core (the critical one): Speech-to-text with word-level timestamps in English, Hebrew, French, Spanish, Portuguese, and Arabic. Hebrew is our primary acceptance bar (fast conversational speech), and note that two of the six languages — Hebrew and Arabic — are right-to-left. Animated, styled captions rendered onto the video — template-based (colors, fonts, word-by-word highlight animation), with correct RTL rendering and bidirectional text handling (e.g., Latin words or numbers inside an RTL sentence). We provide the design specs of ~10 caption styles. Silence/pause removal with 3 pacing levels, and optional removal of repeated takes. Milestone 2 — Polish: Auto punch-in zooms on key moments. Background-music mixing: track by ID from our library, volume 1–100, fade in/out. Milestone 3 — B-rolls: AI-selected stock-footage overlays covering a configurable % of the video. Milestone 4 — Long-form to clips: Split a long video into N best short clips (min/max length params), with face-centered reframing to 9:16. Integration requirements (non-negotiable) REST API: POST /projects accepting ( video, language, captionTemplateId, magicZooms, brolls + %, removeBadTakes, silencePace, music: trackId, volume, fade , webhookUrl ) → async processing → webhook callback with a downloadable MP4 URL. We will share our current API contract so the service is a drop-in swap for our backend. Two ingestion paths, both required: (a) direct upload from our apps — mobile included — so design for large files over flaky connections (resumable/chunked upload, e.g. tus or S3-compatible multipart); (b) pull from a URL we provide (our existing storage). Typical input: 1–10 min vertical videos, files can be large (2GB+). Target turnaround: minutes, not hours. Deployable to our cloud account (we own the infra and the code). Include a processing queue and basic monitoring. You should have Proven work with FFmpeg and programmatic video rendering (Remotion, Motion Canvas, or raw FFmpeg filter graphs). Hands-on experience with ASR (Whisper large-v3, Deepgram, or similar) including word-level alignment; RTL experience (Hebrew/Arabic) is a big plus. Backend + GPU-pipeline experience (Node or Python), API design, webhooks, queues, resumable uploads. Portfolio links to similar video-automation work — required. How we'll evaluate Milestone-based contract. Milestone 1 acceptance: a blind side-by-side against our current provider on 10 real Hebrew videos — caption accuracy, timing, and visual quality judged by us — plus spot checks in the other five languages. We pay per accepted milestone.