Project Name: AuraStyle AI

💡 Inspiration

We noticed a massive gap in the digital beauty and fashion industry: most virtual stylists only focus on either clothing or skincare, and they rarely offer true, clinical-grade personalization. We wanted to build a hyper-personalized, multi-modal AI Stylist and Beauty Lab that not only curates your high-fashion aesthetic through conversational AI but also performs a biometric scan of your face to recommend a clinical skincare routine—all in one seamless, glassmorphic interface.

⚙️ What it does

AuraStyle AI is a dual-engine personal styling and beauty platform:

1. The Persona Builder: Users chat with a high-end AI stylist to define their aesthetic (e.g., Cyberpunk, Minimalist, Haute Couture). The system dynamically generates a customized mood board and curates fashion accessories perfectly matched to their vibe.
2. The Beauty Lab: Users upload a selfie, and the application performs a real-time, pixel-by-pixel biometric skin analysis to detect hydration (moisture), acne levels, and texture. Based on the diagnostic scores, it automatically recommends curated skincare products.

🛠️ How we built it

We architected AuraStyle AI to be incredibly resilient, utilizing a multi-agent architecture and top-tier AI services:

• Frontend & Architecture: Built entirely on Next.js (App Router) with a stunning, highly responsive UI featuring Framer Motion micro-animations and a sleek glassmorphic design system.
• TrueFoundry AI Gateway: We routed our conversational Persona Builder prompts through the TrueFoundry AI Gateway. This allowed us to manage our LLM routing effortlessly and test model resilience, ensuring our AI stylist maintained 100% uptime with sub-second latency.
• TrueFoundry CLI: We utilized the TrueFoundry CLI (tfy) to manage our environments and platform resources programmatically, making our deployment workflow incredibly lightweight and fast.
• Computer Vision & S2S Integration: We integrated the Perfect Corp Skin Analysis engine by building a complex Server-to-Server (S2S) polling architecture. We bypassed standard SDKs to manually authenticate, directly stream raw binary image buffers (Blob natively in Next.js) to their AWS S3 buckets, and asynchronously poll their AI engine for live biometric metrics (Moisture, Acne, Texture).

🏗️ Software Architecture Map

Here is the detailed flow of data powering AuraStyle AI:

flowchart TD
    %% Define styles
    classDef client fill:#0f172a,stroke:#38bdf8,stroke-width:2px,color:#fff
    classDef backend fill:#1e293b,stroke:#818cf8,stroke-width:2px,color:#fff
    classDef external fill:#334155,stroke:#fbbf24,stroke-width:2px,color:#fff

    %% Components
    Client["📱 Next.js Client (React/Browser)\nUI Components & State Management"]:::client

    subgraph "Serverless Backend (Next.js App Router)"
        ChatAPI["/api/chat\n(LLM Prompt Router)"]:::backend
        SkinAPI["/api/skin-analysis\n(S2S File & Task Orchestrator)"]:::backend
    end

    subgraph "External Infrastructure & ML Engines"
        TrueFoundry["🧠 TrueFoundry AI Gateway\n(Model Routing & Failover)"]:::external
        OpenAI["OpenAI gpt-3.5-turbo\n(Persona Generation)"]:::external

        S3["🪣 Amazon S3\n(Temporary Blob Storage)"]:::external
        PerfectCorp["👁️ Perfect Corp AI\n(Computer Vision Engine)"]:::external
    end

    %% Client to Backend routing
    Client -- "Text Prompts (JSON)" --> ChatAPI
    Client -- "FormData (Web Blob)" --> SkinAPI

    %% Persona Builder Data Flow
    ChatAPI -- "Bearer Auth + Config" --> TrueFoundry
    TrueFoundry -- "Proxy Request" --> OpenAI
    OpenAI -. "Text Response" .-> TrueFoundry
    TrueFoundry -. "JSON Payload" .-> ChatAPI
    ChatAPI -. "Styled Response" .-> Client

    %% Beauty Lab Data Flow (3-Step S2S)
    SkinAPI -- "1. Init Upload (Metadata)" --> PerfectCorp
    PerfectCorp -. "Presigned S3 URL" .-> SkinAPI
    SkinAPI -- "2. Stream Raw Binary (PUT)" --> S3
    SkinAPI -- "3. Trigger Task (file_id)" --> PerfectCorp

    %% Polling Loop
    SkinAPI -- "4. Poll Status (1s interval)" --> PerfectCorp
    PerfectCorp -. "Biometric Metrics JSON" .-> SkinAPI
    SkinAPI -. "Parsed Scores" .-> Client

🚧 Challenges we ran into

Integrating a strictly validated, asynchronous computer vision API into a serverless environment was incredibly challenging. The AI skin diagnostic engine required exact binary formats and unpredictable polling times. We overcame this by designing a custom asynchronous polling loop inside our Next.js backend that intercepts any image validation errors (such as poor lighting or lack of facial recognition) and gracefully handles the UX without ever crashing the application.

🏆 Accomplishments that we're proud of

• Successfully combining highly distinct AI models—Large Language Models for stylistic conversation and Computer Vision models for biometric diagnostics—into a single, unified, latency-free user experience.
• Mastering the TrueFoundry AI Gateway for our LLM management, preventing hard-coded model limitations and allowing our architecture to scale efficiently.

📚 What we learned

• The immense value of using the TrueFoundry ecosystem (AI Gateway and CLI) to cleanly orchestrate and manage AI agent deployments instead of relying on direct, fragile API connections.
• Deep architectural knowledge of Server-to-Server binary file streaming and asynchronous task polling in modern JavaScript frameworks.

🚀 What's next for AuraStyle AI

Our next major milestone is to leverage TrueFoundry's full multi-agent orchestration capabilities. We plan to introduce a Voice-Analyzer Agent (inspired by TrueFoundry's multi-agent architectures) to allow users to verbally describe their desired look, having the AI simultaneously analyze their tone and skin to curate the ultimate, hands-free personal styling experience.

Built With

• next.js
• node.js
• pollinations-ai
• react
• remotion
• tailwindcss
• typescript