Kunnskapsassistent (prototype)

An AI tool for knowledge workers to analyze large document repositories.

RAG techniques ensure grounded responses to user queries.

Tech stack:

• Electric Clojure v2
• Typesense v28
• Datahike + Postgres
• OpenAI or Azure OpenAI compatible inference API
• ColBERT reranker

Local development

Global dependencies

Java and Clojure

MacOS, using Homebrew

brew install openjdk@21

NB: be sure to add the necessary symlink as instructed at the end of the previous command

Clojure command line tools:

brew install clojure/tools/clojure

Yarn 4.x.x

corepack enable yarn set version stable

Install project dependencies

yarn install to install Tailwind and other javascript dependencies

yarn build:tailwind:dev to build the css watch and build

Dev build:

• Shell: clj -A:dev -X dev/-main, or repl: (dev/-main)
• http://localhost:8080
• Hot code reloading works: edit -> save -> see app reload in browser

Production build:

clj -X:build:prod build-client
clj -M:prod -m prod

Docker

see Dockerfile

Runtime env variables:

For local development, you can specify values for Postgres, Typesense, Azure OpenAI and ColBERT reranker secrets in the mise.development.local.toml file (gitignored).

In production, specify the secrets in the environment variables of the server.