AI Engineer to Build Self-Hosted AI Avatar Video Platform

I'm looking for an experienced AI engineer / ML engineer / full-stack developer to help build a self-hosted AI avatar video generation platform. I do not want this project to depend on third-party avatar APIs like HeyGen, D-ID, Synthesia, or similar platforms. The goal is to own the core technology stack as much as possible and avoid dependency on companies changing or deprecating their APIs. The platform should allow users to: Upload video footage to create/train a reusable AI avatar Upload an MP3 audio file Generate an MP4 video of the selected avatar speaking that audio Upload a photo and turn it into an animated/talking avatar Save and reuse multiple avatars Potentially scale this into a SaaS product later The ideal approach would be based on: Open-source AI video / talking-head / lip-sync models Self-hosted inference Cloud GPU or dedicated GPU infrastructure Custom training/fine-tuning where needed A full web app for uploading assets, managing avatars, processing jobs, and downloading final videos Architecture that can eventually support multiple users and paid SaaS accounts I am open to using open-source models and infrastructure providers, but I do not want the core product to rely on closed third-party avatar-generation APIs. For the first phase, I am looking for: Technical feasibility assessment Recommended open-source models and architecture MVP roadmap Infrastructure requirements GPU cost estimates Timeline and budget estimate Explanation of the limitations compared to tools like HeyGen Advice on how difficult it would be to turn this into a SaaS platform Please apply only if you have hands-on experience with at least some of the following: Talking-head generation Lip-sync models Video generation models Face animation Voice/audio-driven animation Diffusion models Computer vision PyTorch GPU deployment FastAPI / backend development SaaS architecture In your application, please include: Similar AI video/avatar projects you have built Which open-source models you would consider using Whether you recommend training custom avatars or using reference-based animation Expected quality limitations Estimated cost for an MVP Estimated processing cost per generated video What you would build first to validate the idea Important: I am not looking for a simple wrapper around existing avatar APIs. I want to explore building an independent, self-hosted avatar video generation system where the core generation pipeline is owned and controlled by us.