This repo has been deprecated, please refer to https://github.com/lamini-ai/lamini-sdk
Generate a question answer dataset from any earnings call transcript. Perfect for finetuning LLMs that are experts in current financial data.
Clone this repo
git clone git@github.com:lamini-ai/lamini-earnings-calls.git
Get your lamini API key
Run the script
./generate-qa.sh
This script reads from data/earnings-transcripts.jsonl . You can add your own in a format like this:
{
"date": "Aug 27, 2020, 9:00 p.m. ET",
"exchange": "NASDAQ: BILI",
"q": "2020-Q2",
"ticker": "BILI",
"transcript": "..."
}
{
"date": "Jul 30, 2020, 4:30 p.m. ET",
"exchange": "NYSE: GFF",
"q": "2020-Q3",
"ticker": "GFF",
"transcript": "..."
}
You can also modify the data loading function in lamini_earnings_calls/main.py to fit your needs:
async def load_earnings_calls():
path = "/app/lamini-earnings-calls/data/earnings-transcripts.jsonl"
with jsonlines.open(path) as reader:
for line in reader:
logger.info(f"Loaded earnings call for {line['ticker']}")
yield PromptObject(prompt="", data=line)