The Challenge
A busy auction house processes hundreds of items through its cataloguing system, with every item requiring a clear, accurate, and engaging description before it can be listed for auction.
Writing these descriptions manually is repetitive, time-consuming, and costly. However, simply replacing staff with AI wasn’t an option. Auction descriptions need to be factual, consistent, and reliable, and current AI models can still make mistakes if left unsupervised.
The challenge was to create an AI solution that would assist staff rather than replace them, increasing productivity while maintaining quality control.
The Solution
I developed a bespoke AI-assisted description generator designed specifically for the auction industry’s workflow.
The system allows staff to upload photographs of an item, where an advanced vision-language AI model analyses the images and generates a professionally written product description within seconds.
Rather than producing a final listing automatically, the AI creates a high-quality first draft that staff can quickly review, edit if necessary, and approve before publication. This approach combines the speed of AI with the judgement and expertise of experienced staff.
Intelligent AI Integration
One of the key objectives of the project was ensuring the AI produced practical, consistent descriptions without requiring complex machine learning or expensive model training.
Instead, I refined the system through carefully engineered prompts and workflow design, allowing the model to produce reliable, commercially useful descriptions while remaining flexible enough to improve over time. The result was a solution that was:
- Fast to implement.
- Cost-effective.
- Easy for staff to use.
- Continuously improvable through prompt refinement.
- Designed to integrate into the client’s existing workflow.
Human-in-the-Loop AI
One of the biggest misconceptions surrounding AI is that it can simply replace people. In reality, the best business outcomes often come from combining AI with human expertise.
This system was designed with that philosophy in mind. The AI handles repetitive drafting, while experienced staff provide the final review before publication. This ensures descriptions remain accurate, appropriate, and aligned with the auction house’s quality standards while dramatically reducing the amount of time spent writing every listing from scratch.
The Result
The platform significantly reduced the time required to catalogue general auction items while maintaining quality control.
Instead of beginning every description with a blank page, staff now receive a professionally written draft in seconds, allowing them to focus on reviewing and refining rather than creating content from scratch.
Benefits Delivered
- AI-generated product descriptions from uploaded images.
- Faster cataloguing workflow.
- Reduced repetitive administration.
- Improved staff productivity.
- Consistent description formatting.
- Human review before publication.
- Flexible prompt-driven improvements without retraining AI models.
Why This Project Matters
Artificial Intelligence is most effective when it enhances people rather than attempts to replace them.
This project demonstrates how businesses can integrate AI in a practical, responsible way, automating repetitive tasks while keeping humans in control of quality and decision-making.
Beyond simply connecting to an AI model, the platform was engineered for production use, with built-in reliability features that manage concurrent requests, handle API rate limits gracefully, and support high-volume processing without disrupting staff workflows.
Technology Stack
Backend
- Node.js
- Express.js
Artificial Intelligence
- Qwen2-VL-7B Vision-Language Model
- Replicate API integration
- Prompt engineering for accurate image-based product descriptions
Frontend
- Custom HTML, CSS & JavaScript interface
- Multi-lot worksheet supporting uploads of up to six images per auction item
Workflow & Export
- Automated Excel spreadsheet generation
- Bulk photo ZIP export
- Optimised for high-volume auction cataloguing
Security & Reliability
- Password-protected access using HMAC-signed session authentication
- 24-hour secure session management
- Server-side concurrency control for stable AI processing
- Intelligent request throttling
- Automatic exponential backoff and retry handling for API rate limits
Hosting
- Railway cloud infrastructure