This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

What I Built

CatalogCraft AI is a local-first, AI-powered product listing and cataloging engine designed for independent e-commerce sellers, dropshippers, cataloging freelancers, and small D2C brands selling on platforms such as Amazon, Etsy, Shopify, and Meesho.

The idea is simple: turn a product photo and a few notes into a structured, marketplace-ready product listing without paying per-request cloud AI API fees.

The Pain Point & Side Hustle Utility:

Small online sellers spend hours manually creating product titles, descriptions, SEO keywords, feature bullets, and catalog spreadsheets for their inventory.

This becomes particularly challenging when a seller needs to upload hundreds of products across multiple marketplaces, each with different listing conventions and formatting requirements.

Cloud-based AI tools can help, but repeated API usage can add costs. Uploading unreleased product designs to external services may also raise privacy and confidentiality concerns.

CatalogCraft AI addresses these challenges by processing product images and seller notes locally and generating:

1.Marketplace-tailored product titles adapted to the selected platform.

2.Five structured feature and benefit bullet points for product listings.

3.SEO keywords and search tags to support product discoverability.

4.Meta descriptions intended for search engine listings.

5.Suggested pricing strategies with a price range and rationale.

6.Formatted HTML product descriptions for compatible storefronts.

7.One-click CSV exports for individual products and bulk catalogs.

The application supports both single-product generation and bulk image processing through a simple Streamlit interface.

The long-term side-hustle opportunity is to offer it as a desktop tool, a one-time licensed product, or a subscription-based cataloging service for small merchants and cataloging freelancers.

Demo video

Code

How I Built It

CatalogCraft AI was built using an open-weigth Ai model and local interface tools:

1. Open-Weight Models: 1.Moondream — Vision Model Moondream analyzes uploaded product images and generates textual descriptions of visible product attributes,including product type, color, design, material, and visible branding. Instead of directly generating the entire listing from an image, the application first converts the image into a textual description.

2.Llama 3.2 — Text Generation Model
The text model uses the vision description and seller-provided notes
to generate structured, platform-specific product listings.
It can also generate listings from text-only product specifications
when no image is provided.

2.Local Inference Engine:
1.Powered by Ollama, allowing the models to run entirely
on the user's laptop over a local socket with zero

dependency on third-party cloud endpoints.

3.Structured Reliability:
The application uses Ollama's JSON output mode (format="json") and

Python's json.loads() to parse the generated listing.The intended

listing structure includes titles, bullet points, SEO keywords, meta
descriptions, pricing strategies, and HTML descriptions.

A Pydantic model defines the expected product listing schema, while
Pandas converts generated records into CSV-compatible data.

4.Latency Optimization:
1.Added an image preprocessing pipeline using Pillow that
dynamically scales and optimizes image buffers down to
512px before feeding them into the vision model,
reducing image payload size to help improve inference efficiency.

5.Streamlit Interface and Bulk Processing:

The application includes two primary workflows:

Single Product: Upload an image, enter product notes, generate a listing, review the output, and download a CSV file.

Bulk Batch: Upload multiple product images, generate listings sequentially, inspect the resulting catalog in a table, and download a consolidated master CSV.

Why Does Open Innovation Matter?

Open innovation is what makes this project a viable, scalable business rather than an API bill trap:

1.Zero Marginal Unit Economics:
Closed APIs charge for every image token and output generation. For a small business cataloging 5,000 seasonal items, cloud API bills would destroy product margins. With open-weight models running locally via Ollama, generating 5,000 listings costs ₹0 / $0.

2.Strict Data Privacy & IP Protection:
D2C merchants and craftspeople often prepare catalogs months before product launches. With an open-weight local engine, product images and seller notes are processed locally rather than sent to a third-party cloud AI API for generation.

3.Offline & Edge Capability:
Warehouses and packing facilities often suffer from spotty internet. Because CatalogCraft AI can run locally without an active internet connection after the required models and dependencies have been downloaded, merchants can batch-process physical inventory offline.

4.Freedom from Vendor Lock-In:
Open architectures mean we are never at the mercy of sudden API deprecations, rate limits, or unexpected pricing hikes.

My First User

Tested CatalogCraft AI with a local seller listing casual footwear and handcrafted accessories across Amazon and Meesho. Previously, writing titles, tags, and bullets for a batch of 10 products took them nearly 2 hours of manual typing and search console checking.

Using CatalogCraft AI's batch processor, they generated complete listings and downloaded the ready-to-import CSV file in under 3 minutes. Their direct feedback: "Having the Amazon bullet format and pricing strategy suggested instantly saves us hiring a catalog freelancer."

Prize Categories

Best Use of Render