This is a collection of LLM programs for a personalized AI assistant that is driven by Chainlink’s publicly-available developer resources:
Our goal is to improve the productivity of developers that are building with Chainlink infrastructure. Many developers already use ChatGPT, but this is a general model that (i) often outputs instructions that are out of date, (ii) isn’t specialized towards developing on top of Chainlink.
We use a recent approach to personalizing AI assistants, called in-context retrieval-augmented language models (see research overview here), which has the advantage of citing sources and reducing hallucination (making stuff up).
You can find further details of the project in this doc.
We have scraped various data sources relevent to Chainlink development such as the Chainlink Developer Docs, Chainlink Tags on Stack Overflow and Chainlink Academy. We run this text through the OpenAI embedding model and store it in a vector db. This data can be found in the algovate/data folder e.g. documents.pkl.
We have experimented with a variety of LLM assistants using LLM programming approaches like LLM workflows/chains and agents, with frameworks such as LangChain and Llama Index. These LLM assistants use a variety of retrieval methods (e.g. vector-based retrieval), logic and models (e.g. the new 16k token context window model from OpenAI). These can be found in the algovate/langchain and algovate/llama dirs, and notebooks.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent