llamka

MLOps for building second brain

There is few principles we want to build llamka around

Architecture

• llore - (llamka-core) Tornado App that keep everything running. llore provides api and all other services built on top of it

  • MVP
    • provide abstraction interface over LLM providers (local: ollama; remote: openai, claude;)
    • maintain external document locations/versions and rebuild vector db collections on changes
    • augmentation pipelines and chatbot config
    • completion api to interact with configured models and chatbots
  • later
    • store chatbot interactions in oumi conversation format per user if instructed
    • maintain chatbot library (includes: system prompt, finetuning and augmentation pipelines, data dependencies, and build schedule)
    • maintain conversation datasets and use them for tuning, RAG and testing
    • schedule and run pipelines: finetuning, RAG, chatbot evaluation
    • monitor and run services: databases, queues, proxies, ui (llit)

• llit - streamlit UI

  • MVP
    • Chatbot interaction
  • later
    • Inspect previous conversations or from library
    • Inspect data: sql, document db, vector db
    • Inspect logs from pipelines
    • Configuration: models, chatbot, pipelines

Got API and streamlit endpoints working, almost MVP!!!

Project Docs

For how to install uv and Python, see installation.md.

For development workflows, see development.md.

For instructions on publishing to PyPI, see publishing.md.

This project was built from simple-modern-uv.