Show HN: Building ElectionGPT, a RAG powered chatbot grounded on candidate data
medium.comDisclaimer: I am the co-founder of Neum AI, a data processing platform for vector embedding generation and data retrieval for RAG.
With the upcoming elections 2024 coming up, we wanted to build a chatbot to allow people to interact with presidential candidates. We wanted the chatbot to be as authentic and unbiased as possible regarding each candidates' perspectives. To do this, we based each candidates KB on their own generated information including tweets, website and government plan. We then sprinkled in generally considered unbiased sources like Wikipedia and Ballotpedia that provide additional context. (Some re-ranking being done in the backend to favor candidate generated data sources) We leveraged our core platform (Neum AI) to consolidate the data into individual indexes for each candidate. These indexes are refreshed daily to pull in any new information. (Ex. latest tweets)
You can access the open-source project for the chatbot here: https://github.com/NeumTry/electiongpt-frontend So it was trained on a bunch of lies? Lies or not lies, the point was the train on the authentic message that the candidate wanted to provide. Try to be as unbiased as possible. are you doing any audio / video transcription?