ML/STUDIOSresearch-grade applied AI/ML
ML Studios AI · Portfolio

Research-grade
applied AI & ML.

Computer vision, AI assistants and smart recommendations — engineered with the depth of research and the discipline of production.

Code github.com/mlstudios-ai | Contact anna@mlstudios.ai
Selected work
03 featured · 02 research
SecondSight hazard-detection app in use secondsight · conf 0.98
03 — Computer vision · edge AI

SecondSight

Best Use Case Award iOS · watchOS Prototype

A phone-based assistant that helps blind and low-vision people avoid everyday hazards — spotting obstacles on the ground in real time and warning by sound or a buzz on the phone or Apple Watch, plus on-demand scene description.

Warns in <300msWorks offlineHands-free alerts
Watch demo → Case study → Code
AI stylist · virtual try-on
Raven.
live · raven.mlstudios.ai
02 — AI assistant · fine-tuned model

Raven AI Stylist

Live Multi-agent Virtual try-on

A conversational shopping stylist that gives personalised outfit advice and shows how a look might appear with AI-generated try-on images — powered by a model trained specifically for fashion working alongside a general one.

Adapts to your tasteSee the outfit, not just read itInstant replies
Open live app → Case study →
Grounded RAG · cited answers
Q&A.
live · apraqa.mlstudios.ai
01 — Retrieval AI · regulated domain

APRA Q&A Agent

Live Grounded RAG

A question-answering assistant for dense banking regulation. It answers only from the official standards and always cites the exact source — and, crucially, says "I don't know" instead of guessing when the answer isn't there. In a regulated field, that trustworthiness is the whole point.

Cites standard & clauseRefuses to guess12× more accurate than basic RAG
Open live app → Case study → Code
Research & prototypes
02 from the lab
MRI brain scanroi · hippocampus
R-01 — Medical imaging AI

Medical AI

92%accuracy — ahead of the 87% specialist benchmark

Detecting early Alzheimer's from MRI brain scans. By focusing the model on the region where the disease appears first, it reads scans more accurately than the human specialist benchmark.

Case study →Code
SPARK recommendation systemreward · online
R-02 — Reinforcement learning

SPARK

Livelearns & adapts as tastes change

A recommendation engine that learns from what shoppers actually do — clicks, likes, reviews, purchases — and keeps adapting as their tastes shift, instead of relying on fixed rules that go stale.

Case study →Code
Approach

The work isn't proving the idea works — it's making it something you can deploy and trust.

/01

End to end, owned

One accountable process from raw data to a working product — no messy handoffs, no gaps.

/02

Trusted, not assumed

Automatic checks catch mistakes before users do — and the system is built to admit when it doesn't know.

/03

Real-world minded

Designed to run reliably and affordably, monitored end to end, and quick to update when things change.