Avatar Live Streaming SAAS
Private SaaS for live-streaming workflows, with a web control dashboard, backend orchestration, and AI-assisted routing inside a media-heavy system.
What It Is
Live Avatar Studio is a private SaaS I am building for live-streaming workflows.
The system combines a web dashboard, backend services, a streaming pipeline, and a Windows client that connects into the TikTok streaming workflow.
The Technical Shape
At a high level, the system has a few parts:
- a web dashboard for managing the stream and monitoring playback
- a backend for queueing, orchestration, and APIs
- a streaming pipeline for media playout and delivery
- a Windows client for getting the stream into the TikTok dashboard
- a database layer with vector search
The main stack is Next.js, FastAPI, PostgreSQL + pgvector, FFmpeg, and MediaMTX.
What Makes It Interesting
This is the first project where I had to properly learn streaming concepts instead of treating them like black boxes.
That meant getting more familiar with how RTMP, HLS, and live media pipelines actually work under the hood, then building around the constraints instead of pretending they were just normal web APIs.
- stream control has to feel responsive
- playback has to stay stable
- the Windows client and web stack have to work together
- AI is used as part of the workflow, not as the whole product
That combination makes it a good project for working through real product and engineering tradeoffs.
What I Have Been Working Through
Some of the recurring engineering themes have been:
- keeping stream playback stable while the queue changes underneath it
- keeping videos playing gaplessly and without interrupting the stream
- building a control UI that stays usable during live operation
- using embeddings and similarity search in a practical way
- connecting desktop streaming software with a browser-based control surface
The streaming side has probably been the most educational part. Making video playback feel continuous sounds simple until you actually have to deal with playout, timing, stream handoff, and what happens when the queue changes while a stream is already live.
I like this project because it forces me to think about reliability and workflow, not just features.
Stack
- Next.js
- FastAPI
- PostgreSQL with pgvector
- FFmpeg
- MediaMTX
- Windows client
- RTMP and HLS streaming