Dual-AI Stock Sentiment Analyzer
One model asks smart questions about a company; another judges the mood of the news it finds.
An automated financial-intelligence system that combines two AI models to evaluate market sentiment from real-time news.
Flan-T5 generates context-aware search terms for a company; FinBERT then classifies the sentiment of the resulting articles on a finance-trained three-way scale.
Highlights
- Dynamic query generation instead of static keyword lists.
- Google News RSS aggregation with deduplication.
- GPU-accelerated FinBERT inference; AWS EC2 / SageMaker compatible.
Two models, two jobs
Splitting 'what to search for' (Flan-T5) from 'how does this read' (FinBERT) means each model does what it's good at. The query model keeps the search relevant as news shifts; the sentiment model brings finance-specific language understanding a general model lacks.
Stack
Models
- Flan-T5
- FinBERT
Runtime
- PyTorch
- CUDA
- AWS