Case Study

  - GroupConscience.ai

Project Summery

GroupConscience.ai was developed as a collaborative human in the middle AI trainng loop. The online application gives individual members a chance to contribute and vote on posted truths put forward by fellow recovering alcoholics. In turn, the truths that have the highest acceptance are periodically ingested to improve the AI model. Thus, the responses reflect the conscience of the people that have contributed over time.

This is a bottom-up, rather than top-down approach. At GroupConscience.ai individuals decide which posts to promote rather than companies or other outside influences. In the end the LLM has the abillity to respond in line with the conscience of it's members rather than sounding clinical or having the tone of a Wikipedia article.

Here is a distilled list of the design challenges that were addressed durring the creation of the system.

1. The application must be secure while not using any external API's to handle member data.
2. The responses must be synthesized and returned within 2000ms using a single RTX4090GPU.
3. Member comments that have grater than or equal to a variable number of up votes, must be stored in a vector database.
4. All AI responses must be based on AA literature and or member input. No information outside of these sources can be included in any AI response.
5. Members must be able to star, I.E. save, truths so they are available in their login portal for later viewing.
6. Members must be able to review and edit ther inputs through the member portal.
7. All memeber data must remain Anonymous. No identifiable information can be stored within the system.

Visit GroupConscience.ai