https://github.com/metaskills/unremarkable-vh-rag
https://github.com/metaskills/unremarkable-vh-rag
Last synced: 6 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/metaskills/unremarkable-vh-rag
- Owner: metaskills
- License: mit
- Created: 2024-03-05T12:48:04.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2024-04-03T19:15:32.000Z (over 1 year ago)
- Last Synced: 2025-04-01T17:13:18.688Z (7 months ago)
- Language: JavaScript
- Size: 4.68 MB
- Stars: 0
- Watchers: 2
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE.md
Awesome Lists containing this project
README
# unRemarkable.ai GPT vs. Assistants Demo
Exploring a panel of experts assistants model.
* Mixing Vertical Horizontal RAG Architectures
* Faceted Search, Semantic Retrieval, & Data Analysis
## Setup
This project leverages [Dev Containers](https://containers.dev/) meaning you can open it in any supporting IDE and get started right away. This includes using [VS Code with Dev Containers](https://www.youtube.com/watch?v=b1RavPr_878) which is the recommended approach.
Once opened in your development container, create a `.env.development.local` file with your OpenAI API key:
```
OPENAI_API_KEY=sk-...
```
Now you can run the following commands:
```bash
npm install
npm run db:create
npm run db:index
```
## Demo Commands
This will run the various OpenAI Assistants API demos. The `demo:gpt` command will mimic a Custom GPT's knowledge retrieval. The `demo:rag` will demonstrate the same Q&A sequence with a panel of experts using the Assistants API.
```bash
npm run demo:gpt
npm run demo:rag
```
Images created by any run step such as charts created by code interpreter will be saved to the `./files` directory. These are git ignored. You can use the following environment variables to customize the scripts' behaviors.
- `DEBUG` Set to any value to enable debug 🪲 logging.
- `KNOWLEDGE_FORMAT` Set to process a specific knowledge format. Options: `csv`, `json`, `md`. Defaults to CSV. Only used in the `demo:gpt` script.
## Access OpenSearch
To access the OpenSearch Dashboards hosted in the dev container, use the following URL. Use the `admin` username and `admin` password to log in.
```
http://localhost:5601
```
## Luxury Apparel Data
The data used for this demonstration is from Kaggle.
https://www.kaggle.com/datasets/chitwanmanchanda/luxury-apparel-data
Here are the categories in the dataset. Use these to test your own pre-filtered queries.
* Accessories
* Activewear
* Jackets/Coats
* Jewelry
* Pants
* Shirts
* Shoes
* Suits
* Sweaters
* Underwear and Nightwear
## Notes
Consider 1st tier pass-thru tool data. For example, if `products` responds with `Yes.` is there any guarantee that the orchestrator should echo that message or answer in it's own words? **Should products return raw tool data only or LLM response to said data?**