AI Engineering Capstone · Live Media System
VirtuAI
VirtuAI is a complete media operation behind one virtual creator. Every four hours it finds live AI stories, decides what matters, retrieves a consistent creator voice, creates native content for five platforms, verifies claims and turns human review into evaluation data.
01 · The problem
Why it exists
Generating a post is easy. Deciding what deserves attention, preserving one voice across channels, proving factual claims and learning from editorial rejection are separate engineering problems. VirtuAI connects them inside one accountable operating loop.
02 · Ownership
What I built
I designed and engineered the product, AI pipeline, backend, frontend, safety gates, evaluation harness, publishing workflow and operating proof for my final year Artificial Intelligence Engineering capstone.
03 · System
How it works
The system combines live signal discovery, editorial ranking, Style RAG, multi-agent orchestration, entity and claim grounding, gated publishing, observability and a human review loop. Human decisions become controlled evaluation labels rather than disappearing after approval.
04 · Evidence
What is proven
- Five publishing channels
- A four hour signal cycle
- 437 controlled human labels
- 150 passing tests
- A dataset audit that found 29 percent inconsistency in the original reward labels