📺 Watch the full tutorial on YouTube

🌐 Liquid d1 Multimodal Triage Engine

Liquid d1 Multimodal Support Triage is a zero-token decision engine that analyzes customer complaint text and damage photos simultaneously in a single forward pass without conversational delay.

⚡ How It Works

Step 1: Input	Step 2: AI Action	Step 3: Result
Customer submits text message and broken item photo	Model evaluates routing, severity, and photo damage simultaneously	Calibrated JSON decision, genuine claim verification, and response

📦 Features

• ⚡ Zero-Token Latency: Delivers instantaneous decisions without waiting for autoregressive text stream generation.
• 🖼️ Multimodal Damage Assessment: Cross-checks photo visual evidence against customer text claims in one single forward pass.
• 🎯 Automated Team Routing: Accurately routes claims to Replacements, Fragile Claims, or Safety Escalations.
• 🛡️ Genuineness & Fraud Screening: Validates physical breakage patterns to filter fraudulent or unrelated photo submissions.
• ⏱️ Sub-20ms Processing: Records execution latency per request for high-throughput enterprise queues.

🛠️ Tech Stack

• Model: Liquid AI d1-omni-600M / d1-3B
• Framework: Python 3.10+, PIL (Pillow), PyTorch
• GGUF Runtime: llama.cpp / Ollama (llama-server) with mmproj-model-f16.gguf

🚀 Setup & Installation

Install dependencies with a single command:

pip install "transformers>=5.15" torch torchvision pillow requests

Download Model Checkpoints

Download the official 8-bit quantized weights and vision projector into the models/ directory:

hf download LiquidAI/d1-omni-600M-GGUF d1-omni-600M-Q8_0.gguf --local-dir models
hf download LiquidAI/d1-omni-600M-GGUF mmproj-model-f16.gguf --local-dir models

Register with Ollama

Register the local checkpoint in Ollama using the provided Modelfile:

ollama create liquid-d1 -f Modelfile

Alternatively, serve via llama-server with multimodal projector enabled:

llama-server -m models/d1-omni-600M-Q8_0.gguf --mmproj models/mmproj-model-f16.gguf -b 4096 -ub 4096 --port 8080

💻 Running the Application

Run the multimodal triage pipeline across all customer complaints and damage photos:

python main.py

The script evaluates all 5 customer cases, verifies visual damage, calculates latency in milliseconds, generates customer replies, and exports results to outputs/outputs.md.

📁 File Structure

Liquid d1/
├── Modelfile                 # Ollama model registration
├── README.md                 # Setup and architecture documentation
├── inputs/                   # Customer complaints & damage photos
│   ├── case_1_plates.png     # Smashed dinner plates
│   ├── case_2_cup.png        # Broken ceramic mug
│   ├── case_3_cooktop.png    # Cracked induction cooktop
│   ├── case_4_dish.png       # Split porcelain platter
│   ├── case_5_luggage.png    # Crushed suitcase
│   └── complaints.json       # Customer complaint texts & order metadata
├── main.py                   # Production multimodal triage engine
└── models/                   # GGUF weights & multimodal projector
    ├── d1-omni-600M-Q8_0.gguf
    └── mmproj-model-f16.gguf

🎯 5 Use Cases

1. E-Commerce Damage Claims: Instant verification of broken merchandise photos for automated zero-touch refunds.
2. Airline Baggage Damage Triage: Rapid evaluation of luggage damage photos and compensation routing.
3. Hazardous Appliance Screening: Automated escalation of cracked electrical cooktops and shattered glass items.
4. Logistics Courier Dispute Resolution: Cross-verifying crushed packaging vs intact items for shipping insurance.
5. Restaurant & Catering Supply: High-throughput batch processing of transit-damaged tableware.

🔮 5 Future Features

1. Web Dashboard: Interactive single-page UI for live drag-and-drop customer ticket submissions.
2. Optical Character Recognition: Automatic invoice and serial number reading from parcel labels.
3. Automated Warehouse Restock: Direct integration with ERP inventory for immediate replacement dispatch.
4. Fraud Ring Correlation: Image hash comparison across multiple accounts to detect recycled damage photos.
5. Customer SMS Automation: Instant dispatch of claim resolution updates to customer mobile devices.

---

🏷️ Keywords & Tags

Liquid AI Liquid d1 d1-omni Multimodal AI Zero-Token Architecture Edge AI Ollama GGUF llama.cpp Customer Support AI Damage Detection Vision AI Real-Time Inference