itemLens

Inventory management (for at home). There are many like it, but this one is mine.

The primary use is: Do I have that, where the heck is it? and Why did I buy it?

I am no fan of data-entry, so, adding new products/items should be as automated as possible (using any device). Most of the effort went into this bit using machine learning (LLMs, object classification, OCR, etc).

How to use

To add a product, grab your phone and take a picture of the product. Then scan the QR-code on the container to place it in. (Note: this is the goal. Getting there.)

That's it.

If feeling particularly ambitious on a day, you can also:

• take a picture of an invoice/receipt (itemLens will use image classification/OCR/LLM to get the juicy bits)
• add additional photos (using camera or just paste in links)
• scan QR-codes containing URLs to relevant documents
• paste in a list of attributes (weight/color/size/etc)
• add tags, amount, description, etc (but then you are obviously very ambitious as it might require typing)

Features

• Paste-parser for key-value-pairs
• QR-code reading (server and client)
• Optical Character Reading (OCR)
• Image classification (Blip)
• LLM Summaries (Llama3, Groq)
• Invoice/receipt data extraction
• Download-and-store documents (link-rot no more)
• Image processing (background removal, thumbnail, etc)
• Color extraction
• Multiple inventories (i.e. one for shoes, another for clothes, and yet another for electronics)
• ...and more

Info how I use it

• TODO:
  • Which label printer
  • Which cabinets
  • Pictures of containers
  • Firefox QR-code generator for current link
  • I use it for electronics/components
  • Which fields I actually fill in
  • How I search for related links

Third parties

I really dislike it when I have to register for some 3rd party services to try out some software, therefore, that is all voluntary. Set the flag NO_THIRD_PARTY_SERVICES to true in .env and you can use it all -- but adding new products will be more work.

About third party services and incurred costs

Goal is: No fixed cost / month -- only pay for use

Development

Stack

SvelteKit 2, PWA, Prisma, SQLite, Tailwind CSS, TypeScript, LLMs + various ML models.

Installing

Get it:

npx degit romland/itemLens itemLens

Create some directories

mkdir `static/images`
mkdir `static/images/u`

Fill in your details in .env

cp .env.example .env

Run for development:

cd sveltekit-starter
npm install
npm run dev

Database migration:

npx prisma migrate dev --name init

Database seeding:

npx prisma db seed

Use default user:

Username: admin
Password: password

Build for production

npm run build

External services

Notes to self for now:
Start the docker containers in Ubuntu VM if they are not running with: containers-start.sh

On the Jetson on my desktop: For object classification: ...

Prototype-y

• Right now the code very JavaScripty while it should be TypeScripty. A lot of it is still very much a prototype.

TODO / notes

• make a multi-"screen" wizard for adding new items on smaller devices (from desktop single screen is good)
• consider: Is it faster to do a quick pre-check on the Jetson to see if there is a QR code in image?
• Investigate how fast classification inference can run on a recent RasPi (using OpenCL)
• TODO fields when adding items: inventory Inventory? @relation(fields: [inventoryId], references: [id]) inventoryId Int? usage InUse[]
• Need some thinking about logic to take valuable data from photos and apply it to items for searching, right now we search all.
• autostart containers if they are not running (if Windows, need to start in WSL)
• fetching interesting links (especially documentation/specs) for newly items should also be automatic

TODO README:

Screenshot(s), logo, video(s)