[[The main challenge for AI collaborators is lack of context and specific knowledge about particular use cases]]. This greatly limits their ability to be useful collaborators. [[Context as a dialogic concept]] for my thesis. I am currently implementing (along with my collaborators at Humans Plus AI Studio) two AI collaborators for two use cases: 1. An interior design collaborator that can help designers conceptually brainstorm and also source materials. 2. A law collaborator that helps lawyers modify contracts at scale, and also help them search large legal databases conversationally. Both of these involve being able to learn and access existing knowledge bases. The design collaborator needs knowledge about the interior design's studio knowledge base, and also learn from interactions with them. The law collaborator needs to know legal codes and also potentially learn from interactions. **In order to build this, we need to learn/solve the following.** 1. [[Learn Langchain]]. I am a noobie at this. 2. Knowledge retrieval: vector database integration for knowledge retrieval 3. Conversational learning: rememeber previous conversations and intelligently store in long term memory (also in vector databases) 4. Action performance: getting the agent to take actions 1. This involves outputting prompts in json formats useful for taking actions, or having agents call functions 5. Access to internet (?) 1. For example, a collaborator I am building should be able to help designers source materials, which requires querying the internet. This is part of context awareness but is more detailed. Also involves a level of agentic behavior. 6. Intelligently managing all the above. 1. It is not trivial to have the agent decide when to consult a vector database, take an action, consult the internet 7. Interface: implementation of a chat interface Important resources: https://python.langchain.com/docs/use_cases/question_answering/