LLM News Scoring and Entity Extraction
10 Academy · 2022 · Academic project · Individual
An LLM evaluation pipeline that scores news for topic relevance and extracts entities from job descriptions, served over FastAPI with DVC-versioned datasets.
- LLMs
- Cohere
- FastAPI
- DVC
- Prompt engineering
Overview
Prompt engineering and fine-tuning on Cohere (and GPT-3) models for two tasks: scoring news items 0 to 10 for topic relevance, and pulling structured entities out of job descriptions. The pipeline keeps prompts modular so providers and topics can be swapped without code changes, and versions its datasets with DVC.