Services
Divisionbyzero Oy specializes in data processing and generative AI solutions. We are not AWS/Azure/GCP lobbyists and don't generally offer expensive and complicated enterprise "reference architectures". Instead, we prefer to utilise state-of-the-art open source solutions and cost effective cloud platforms and sometimes on-prem hardware.
An example: duunirobotti.com
Duunirobotti.com is a SaaS service that stores and analyses all Finnish job postings and provides AI assisted cover letter writing for end users. The service is 100% deployed on Hetzner cloud VM's using open source software.
The service consists of two sub systems: data processing and web application.
Data processing uses a python based logic that is executed as a Prefect workflow. A nightly job fetches all new job postings, extracts information using LLM and generates embeddings for semantic search. Raw data is loaded to a postgres database and DBT models are then used to transform and create tables needed for the web application to efficiently search and display the job postings.
Web application backend is built in Go. With the help of HTMX library the whole frontend is server side rendered. Backend handles the complicated search by building SQL queries optimized for postgres full text- and vector -search capabilities. A custom task queuing -engine is also built into the Go backend. It executes long lasting LLM queries on the background, like users generating new cover letter -versions.
Infrastructure is deployed to two affordable Hetzner VMs where all components of the system are run as services in K3S kubernetes environment. Access to web application is proxied via Cloudflare to shield the service from malicious activity.
AI tasks are mostly executed on on-prem servers using cost effective AMD and Nvidia hardware, open source software and open weight LLM models.