🧠AI-projects

AI, LLM & ML projects that I am working for various reasons. Mostly to get a better understanding of the capabilities, and with all the releases of new things, see what kind of applications I can built in my free time.

Wellness+

I created this project as a experiment and proof-of-concept how one could very quickly put together a directory and service listing website with a lot of content with AI & LLMs. The idea behind this was that one can come up with a basic topic; in this example wellness & health travel to Thailand, and only spend a few weekends to get from point zero to a full website.

Tech Stack

Tech Stack

• Backend: Django & Python

• Frontend: AlpineJS, HTMX, CSS, Bootstrap 5.3

Providers

• FAISS as Vector store

• Anthropic + OpenAI embeddings

• Google Maps API

• Google Business API

• Telegram API

• Railway (hosting)

AI & Content

• Business Profile: I used the Google business API as well as AI & LLMs to create a vector database profile of each business which was used for both question answering about a specific business as well as content generation.

• Content & SEO Generation: It took me several days to come up with Categories, and I used AI & LLM + the Google Business API to generate over 80 locations within 4 cities matching these categories. This in total generated more than 100 pages of context; each page had its SEO (titles, meta, description, og, etc) generated dynamically by an LLM.

• Category, City, Service & Business chat: Having very diverse information about businesses within a specific region, I created a chat through which users could pose standard questions like: "What is the best Fish Spa in Bangkok" and get returned information through RAG on several entries.

The content is segmented in various ways

1. A very large vector database with 60k vectors that AI can access

2. A smaller database with summarized data less than 6000 vectors.

3. And a vector database per location/business (100-1000 vectors)

Depending on the need, we logically determine which database needs to be used

1. Writing content: Large Database

2. Updating information (SEO & Descriptions): Smaller Database

3. Answering questions about businesses: Location Database

4. Finding a list of relevant businesses: Location + Smaller Database

AI Concierge recommender

Tech stack: Flask - HTML, HTMX, AlpineJS, Bootstrap 5.3, CSS For a proof-of-concept demo I worked on a hospitality AI concierge recommender and Chatbot. This recommender gets the users current location and a list of curated locations within a hospitality platform. On the fly, the distance (in meters) as well as the walking duration (in minutes) is calculated.

There are then two options, either the user asks the Chatbot questions about the various places, or he highlights a place, gets the route there and can ask the AI chatbot for more information about this particular place.

This was designed, built and developed over the course of 3 days and is built from the ground up with Flask, HTML, HTMX, CSS & JavaScript. The API that is being used is the Google location API and there are two options, either a local LLM served through LM-Studio or an OpenAI model like GPT3 or GPT4o.

The reason for working on this was two-fold.

1). I saw a interesting use-case with a company that I interviewed with and wanted to explore if I could develop this. 2). Really enjoying HTMX + Flask, I am looking become even more proficient with them.