Show HN: Avatar – Talk to people modeled on their tweets

[SITZ]

Stumbling across an interesting person online and seeing decades of their writing history is always daunting to dig through to discover key ideas.

While standalone foundational models can dig through most popular writings available on open web, there is long tail that needs individual specific models to synthesize with LLMs. Starting with twitter, Avatar intends to create a digital avatar for individuals while aiming to be factually correct. Please Try ~50 interesting accounts on twitter IMO and let me know of any feedback on its utility.

RAG stack components: LLM (gpt-3.5-turbo), Embeddings (text-embedding-ada-002), Vector index (Chroma)

More details: https://twitter-avatar.com

[cdev823]

Really interesting idea! Avatar could be very useful for finding out what people have said about a certain topic in the past. Linking to the original tweets is also really helpful.

Can't wait to see where you take this in the future!