Build video-clip sourcing, identification and editing platform (Python/Node, Docker)

Overview: We run a small content team that finds short movie/TV clips on Instagram, locates the scene in our own full-length source files, edits it into a vertical captioned reel, and publishes it. We have a detailed spec in two phases and want a fixed-price quote for EACH phase separately, plus a total. PHASE 1 — core platform 1. Sourcing - Pull Instagram reels from a list of accounts and hashtags on a schedule, using a scraping API (Apify or similar). - Store post, metrics and thumbnail; dedupe near-identical videos with perceptual hashing. - Daily cost cap on scraping spend. 2. Identification - For each reel, identify the film or TV show, and for TV the season and episode. - Match the reel's dialogue against subtitle text (OpenSubtitles) to confirm the episode and the moment, with picture matching as fallback. - Show title and start–end timestamp on the reel card; "Find it again" and manual override. 3. Auto-edit and render - Cut the matched scene from our source file (cloud storage), overlay captions/subtitles, export vertical 9:16 with ffmpeg. - Finished videos kept in a library. 4. Review dashboard (web app) - Pipeline stages: To review → To record → To caption → Ready to post → Posted, with counts per stage, approve/skip, client filter. - Per-operator queues; admin sees everyone. - Work log per operator — clock in / clock out and a record of what each person reviewed, recorded, captioned and posted. - Activity feed explaining what the system did and why. - Pages: Reels, Library, Imports, Clients, Analytics, Operators, Inbox, Settings. Team accounts with roles. 5. Deployment - Docker Compose on one dedicated Linux server: Postgres, API, workers, gateway, logging, nightly backups, automated tests. PHASE 2 — advanced editing and precision 6. Multi-scene editing - Build one reel from several scenes across different films or episodes: pick scenes, order them, trim each, auto-join with transitions, captions across the whole cut. 7. Exact-moment identification - Frame-accurate start/end timestamp in the source file, not just the approximate scene; confidence score; target zero confidently-wrong results on a labelled test set we provide. 8. Music - Option to add a music track to a reel inside the app: upload or pick from a library, trim, set volume and fade under the dialogue, mixed in at render. Stack: Python 3.12 and Node/TypeScript, Postgres, Docker, ffmpeg, LLM APIs for captioning (OpenRouter/Anthropic), TMDB + OpenSubtitles, Whisper-class speech-to-text. Scale reference: expect roughly 60–80k lines across several services with full test coverage for Phase 1. Target 90%+ correct identification on the first result. What we need from you - Fixed price and timeline for Phase 1. - Fixed price and timeline for Phase 2. - Total, and your hourly rate for maintenance and changes after delivery. - All code, accounts and IP transfer to us on payment. - A short note on anything similar you've built. Timeline: please propose. We expect an experienced developer using modern AI-assisted tooling to deliver Phase 1 in roughly 4–6 weeks; tell us if you think that's wrong and why.