Flight deck
Making sense of what a system sees.
Inverse-LLaVA
Rethinking how vision and language meet, with no alignment pre-training and 45% fewer training samples.
Read case study
Applied scientist / Research & engineering
From research
to working AI.
I’m Xuhui. I work across multimodal learning, recommendation systems, and the infrastructure that brings models into use.
Inside the explorer
A system of ideas.
Four connected disciplines. Explore the work behind each one.
Crew cabin
Building systems around people.
Treverse Recsys
Personalized recommendations, from temporal graph learning to real-time serving and controlled experiments.
Read case study
Service bay
Giving intelligence a way to interact.
AI for Negotiation
Turning negotiation transcripts into research data, reducing annotation cost from $5,000 to $3 per transcript.
Read case study
Propulsion assembly
The infrastructure that brings models into use.
FORGE
A shared ML platform connecting training, deployment, and observability across cloud and edge workloads.
Read case study
02 / How I work
Research and production,
in one loop.
A useful model is only part of the story. I follow the question through evaluation, deployment, and what the system teaches us next.
Method / 05
Each stage opens the project that best demonstrates it.
01 / 05
Open project
Frame the question.
Ancient Mortars structures roughly 10 million particle images for comparison.
02 / 05
Open project
Fit the representation.
Inverse-LLaVA maps text into visual space inside the language model.
03 / 05
Open project
Make the comparison legible.
Negotiation transcript coding reports a cost drop from more than $5,000 to $3.
04 / 05
Open project
Publish a reviewable release.
FORGE defines six contracts across distinct ML runtimes.
05 / 05
Open project
Return field evidence.
Treverse joins controlled exposure with reproducible learning and serving.
03 / Experience