🌟 Style Agent — Hyper-Personalised AI Fashion Stylist

An end-to-end AI agent pipeline that analyses your body type, skin undertone, occasion, vibe, and budget to generate three complete, coherent outfits — each with live Google Shopping links, colour-matched jewellery, and stylist-quality notes.

<br>

📸 What It Does

Style Agent acts as your personal AI stylist. You fill in a few details on the left panel, click Generate My Outfits, and within seconds receive:

• 3 complete outfit looks — clothing, footwear, bag, and matching jewellery
• Outfits correct for your gender (Women / Men), vibe (Ethnic, Modern, Boho…), and occasion (Wedding, Office, Brunch…)
• HSL-based colour palette — three harmonious suggestions (complementary, triadic, split-complementary)
• Real shopping links — 3 clickable pill buttons per item opening Google Shopping, Myntra, and Ajio
• Formality scoring — no sneakers with a bridal lehenga
• AI Chat Stylist — a floating ChatGPT-style dialog powered by Ollama llama3

<br>

🏗️ Architecture

run.py
  └─ LangGraph workflow (workflow/langgraph_state.py)
       ├─ Node 1: PersonaAgent         ← builds user style profile from DB history
       ├─ Node 2: ColourEngineAgent    ← HSL palette math, hex→colour family mapping
       ├─ Node 3: TrendScoutAgent      ← occasion & vibe trend analysis
       ├─ Node 4: WardrobeArchitectAgent ← 5-tier DB query, outfit assembly
       └─ Node 5: JewelleryAgent       ← matches jewellery to skin tone & outfit

All agents run in a LangGraph state machine. The CrewAI crew (workflow/crewai_crew.py) wraps them in a multi-agent task collaboration layer.

<br>

✨ Key Features

| Feature | Detail | |---|---| | 5-Tier Colour Matching | colour_family (warm/cool/neutral/earth/pastel/jewel) fallback so the DB always yields results | | Dense Inventory | 1,185 auto-generated items — Women, Men, Unisex × 17 colours × all vibes | | Formality Scoring | Every occasion scored 1-5; footwear & bag picked to match | | Gender Filter | 👩 Women / 👨 Men toggle routes to correct Indian garment categories | | Indian Ethnic Garments | Lehenga, Sharara, Gharara, Anarkali, Saree, Sherwani, Bandhgala, Kurta, Mojari | | Outfit Coherence Validator | Flags mismatches (sneakers with lehenga, missing dupatta) | | Google Shopping Links | 3 pill buttons per item (Google Shopping · Myntra · Ajio) — always work | | Match Transparency | Pale yellow notice if colour was approximated (explains tier used) | | AI Chat Stylist | Floating ChatGPT-style dialog driven by Ollama llama3 | | Hex Colour Picker | 24-swatch grid + OS colour wheel + manual hex input | | Vibe Tiles | 8 vibes with accent colours and emoji — click to select |

<br>

📁 Project Structure

StyleAgentRetailAnalyst/
│
├── run.py                        # 🚀 Entry point — python run.py
│
├── agents/
│   ├── persona_agent.py          # Agent 1: user style profile
│   ├── colour_engine_agent.py    # Agent 2: palette math + hex→family
│   ├── trend_scout_agent.py      # Agent 3: vibe & occasion trends
│   ├── wardrobe_architect_agent.py # Agent 4: 5-tier DB query + outfit assembly
│   └── jewellery_agent.py        # Agent 5: jewellery matching
│
├── workflow/
│   ├── langgraph_state.py        # LangGraph state machine (agent pipeline)
│   └── crewai_crew.py            # CrewAI multi-agent task crew
│
├── database/
│   ├── setup_database.py         # Creates + seeds all 6 SQLite tables
│   ├── inventory.db              # The live SQLite database (auto-created)
│   └── sql_queries.py            # Query helpers
│
├── gui/
│   └── tkinter_app.py            # Full Tkinter GUI (1,400+ lines)
│
├── scraper/
│   └── live_link_scraper.py      # Builds Google Shopping / Myntra / Ajio URLs
│
├── requirements.txt
└── setup_guide.txt

<br>

🚀 Quick Start

Prerequisites

• Python 3.10+
• Ollama installed and running locally (for AI chat only)

1. Install dependencies

pip install -r requirements.txt

2. Pull the Ollama model (for AI chat)

ollama pull llama3

3. Run the app

python run.py

The database is created automatically on first run — no manual setup needed.

<br>

🧰 Tech Stack

| Layer | Technology | |---|---| | Language | Python 3.10+ | | GUI | Tkinter (built-in) | | Database | SQLite via sqlite3 (built-in) | | Agent Orchestration | LangGraph + CrewAI | | Local LLM | Ollama — llama3 | | Colour Math | colorsys (built-in) — HSL/HSV analysis | | HTTP | requests + beautifulsoup4 (trend data) | | Shopping Links | Google Shopping URL construction (no scraping) |

<br>

🗃️ Database Schema

The SQLite database (database/inventory.db) contains 6 tables:

| Table | Purpose | Rows | |---|---|---| | current_inventory | Full fashion catalogue — programmatically generated | 1,185 | | jewellery_inventory | Jewellery pieces matched to skin tone | 30 | | user_profile | Stored style preferences | 1 (sample) | | purchase_history | Past purchases for persona analysis | seeded | | browsing_logs | Viewed items for personalisation | seeded | | outfit_history | Generated outfits (saved by app) | starts empty |

current_inventory key columns

| Column | Type | Description | |---|---|---| | colour_family | TEXT | warm / cool / neutral / earth / pastel / jewel | | gender | TEXT | Women / Men / Unisex | | formality_score | INTEGER | 1 (casual) → 5 (black tie) | | vibe_tags | TEXT | comma-separated e.g. "ethnic,classic" | | occasion_tags | TEXT | comma-separated e.g. "wedding,sangeet,festive" | | category | TEXT | lehenga / saree / sharara / kurta_pyjama / top / bottom / footwear / bag … |

<br>

🤖 Agent Pipeline Detail

Agent 1 — PersonaAgent

Reads user_profile, purchase_history, and browsing_logs to build a style persona. Outputs comfort level, brand tier affinity, and risk score.

Agent 2 — ColourEngineAgent

Uses colorsys.rgb_to_hsv() to map any hex code to one of 6 colour families, then generates 3 harmonious palettes (complementary, triadic, split-complementary). New in v5: get_search_colours_for_hex() for DB-compatible colour lookups.

Agent 3 — TrendScoutAgent

Analyses the occasion and vibe to output relevant trend keywords, silhouette guidance, and fabric suggestions. Checks live trend data where available.

Agent 4 — WardrobeArchitectAgent (core)

Builds 3 complete outfits using a 5-tier SQL fallback system:

1. colour_family + vibe + occasion + gender + formality ±1
2. colour_family + vibe + gender
3. colour_family + gender
4. gender + formality
5. gender + price (guaranteed result)

Also runs the OutfitCoherenceValidator to catch formality mismatches.

Agent 5 — JewelleryAgent

Picks earrings, necklace, bangles, and optional maang tikka based on the outfit's metal tone (gold for warm skin, silver for cool) and occasion formality.

<br>

📈 v5 Upgrades (7-Fix Series)

| # | Fix | Impact | |---|---|---| | 6 | Dense inventory generator | 1,185 items from 70 templates × 17 colours | | 1 | Colour family matching | No more "0 results" — hex → family → 5-tier fallback | | 7 | Formality scoring | Correct footwear/bags for every occasion type | | 4 | Outfit coherence validator | Flags sneakers-with-lehenga, missing dupattas | | 3 | Gender filter | Correct Indian garments for Women vs Men | | 2 | Google Shopping links | 3 pill buttons per item — always clickable | | 5 | Match transparency messages | Explains any colour approximation to user |

<br>

📄 License

MIT — free to use, modify, and distribute.

---

Built with ❤️ by Chitra Kulkarni — GitHub