EarningsCall-LLM is a specialized language model fine-tuned specifically for analyzing earnings call transcripts. Built on Meta's Llama 3.1 8B architecture using LoRA adaptation, it is designed to extract the financial insights that matter most to investors and analysts from the unique format of quarterly earnings calls.
The model was trained on over 10,000 earnings call transcripts paired with corresponding analyst reports, covering multiple quarters and industries. This specialized training corpus teaches the model to identify forward guidance language, distinguish between scripted remarks and spontaneous Q&A responses, and extract key financial metrics and performance indicators that are often buried in conversational transcript text.
For hedge funds, investment analysts, and asset managers who process hundreds of earnings calls each quarter, EarningsCall-LLM offers a significant efficiency advantage over general-purpose LLMs. Its MIT license allows free use in research and trading workflows, and its focused training on the earnings call format means it consistently outperforms larger general models on tasks like forward guidance extraction, executive sentiment analysis, and automated earnings recap generation.