TestFlight
AI mobile product
AURA - AI Personal Stylist
TestFlight release / active development
A mobile AI styling platform that turns a real wardrobe into personalized outfit recommendations using wardrobe embeddings, vector retrieval, structured agents, and long-term style feedback.
Role and scope
Sole engineer across product design, React Native app, Firebase backend, Cloud Functions, retrieval, recommendation logic, LangGraph orchestration, testing, and beta preparation.
React NativeExpoTypeScriptFirebaseFirestoreCloud FunctionsOpenAILangGraphVector Search
10 nodes
Agent workflow
23 golden cases
Eval harness
0 hallucinated items
Closet validity
Across the current eval suite
Product proof
Real app surfaces
Case study
What this shows
AURA is the portfolio's flagship product: a complete mobile AI styling system that connects wardrobe ingestion, retrieval, agent reasoning, recommendation generation, planning, and feedback loops.
Problem
Personal styling apps often ignore the user's real closet. AURA keeps recommendations grounded in uploaded wardrobe items, saved looks, calendar context, and feedback history.
Problem
Most outfit tools recommend generic looks. AURA is built around closet grounding: recommendations should come from pieces the user actually owns, while still adapting to taste, history, and planning context.
Product experience
The mobile app covers wardrobe upload, AI stylist chat, outfit cards, saved looks, calendar planning, wear tracking, wardrobe insights, and personalized shopping logic.
System architecture
React Native and Expo power the mobile interface. Firebase Authentication, Firestore, Cloud Functions, and Storage provide the backend. OpenAI and LangGraph coordinate the stylist agent and structured recommendation workflows.
Wardrobe intelligence pipeline
Uploaded pieces become structured closet items and retrieval context. Embeddings, category normalization, style memory, and feedback signals help AURA choose relevant items instead of treating every closet piece equally.
Agent workflow
The agent classifies intent, retrieves style memory, retrieves wardrobe context, generates outfits, validates closet membership, repairs invalid output when needed, explains choices, and routes feedback back into personalization.
Production engineering
The project includes TestFlight distribution, privacy controls, account deletion, rate limiting, AI usage gating, fallbacks, request timeouts, and TestFlight-ready QA documentation.
Challenges and decisions
The hardest design choice was prioritizing reliable closet-grounded output over open-ended styling creativity. AURA uses validation, item IDs, role canonicalization, and evaluation gates to keep recommendations grounded.
Current status and next steps
AURA is in active development with a TestFlight release path. Next work focuses on improving personalization loops, expanding shopping recommendation quality, and continuing reliability work around expensive AI operations.
Interested in the implementation?