/OpenLibreAI-android

OpenLibreAI : Open source LLM and generative AI models

Primary LanguageKotlinMIT LicenseMIT

OpenLibreAI Android

Phase One

In phase one of the OpenLibreAI project, the plan is to implement the OpenLibreAI Android app, OpenLibreAI iOS app and the OpenLibreStudio web app

OpenLibreAI app is an AI Assistant and AI Agent which interacts over API interface with various LLMs

Overview

The world needs fine-tuned models for domain specific use-cases to solve domain-specific problems such as image analysis, video analysis, and vector database data analysis.

The inputs to the OpenLibreAI models are raw data.

The output of the OpenLibreAI API (which wrap the OpenLibreAI models) are generated text, images, predictions, summaries and other value-add artifacts.

Mission

OpenLibreAI is a free and open source platform for AI research and development. It is a community-driven project that aims to provide a platform for AI researchers and developers to collaborate and share their work.

What the world needs now is an open source AI stack / cloud platform / application platform.

OpenLibreAI is an open source attempt to collect, deploy, and productionize open source models that are state of the art right now (StabilityAI, etc) as well as the ML models for specific use-cases on HuggingFace.

OpenLibreAI will deploy them under the open source OpenLibreAI platform.

OpenLibreAI's models would be trained and fine-tuned for domain-specific use-cases.

For example, OpenLibre can use skin cancer data and images to train the OpenLibreAI-Carcinoma model that can be deployed as an API and OpenLibreAI gets to charge for the API calls.

OpenLibreAI will also have a marketplace for AI models and APIs.

Workflow

OpenLibreAI will have a workflow that is similar to the following:

  • OpenLibreAI models will need to be trained on good data

  • One of the modules of OpenLibre will be the training data ingestion and annotation module

  • The training data ingestion and annotation module will take a dump of any docs, spreadsheets and csv files and label it for use in training

  • The ingestion module will also need to be domain specific, so there will be different modules for different domains and potentially also different use-cases

  • For example, the cancer prediction moel will need labeled carcinoma imagery

  • Video annotation modules can ingest videos of various events and summarize them, as well as create described-video from them

  • Once one has training data, the LibreAI models can be fine-tuned for all the various use-cases

  • The OpenLibreAI APi then directs the API calls to the specific OpenLibreAI deployed models to do either prediction, summarization, or generative tasks

  • OpenLibreAI has three main categories of capabilities - generative, predictive, and summarization

  • OpenLibreAI can expose these capabilities as paid APIs as well as a reference application that is deployed as a paid SaaS

  • Although the models are being open sourced, the OpenLibreAI monetization model is charging enterprise customers for the API usage of the deployed models (including vector embeddings and context searches)

Vision

  • OpenLibreAI is first and foremost an open source community that believes in deployment of responsible AI for the collective elevation of humanity

  • One of the business models for OpenLibreAI will be to charge for the cleaned training data

OpenLibreAI Collaboration Model and Structure

OpenLibreAI is structured as an open source collective and devs are compensated in proportional-accrued-value-impact-dollars (PAVIDs). PAVIDs are credits, which are pegged to USD so that 1 PAVID = 1 USD.