Financial Predictive Analytics from Unstructured Data

I need help turning a sizable collection of unstructured finance-related text and document feeds into an accurate, production-ready predictive model. My goal is to spot patterns that reliably forecast the target variables I will share once we start (think risk scores and market-driven KPIs). The raw inputs arrive as news articles, earnings transcripts, and PDF reports, so the first step is a solid NLP-centric pipeline that cleans, tokenises, and extracts meaningful features. I work in a regulated environment, which means the code must be reproducible, well-commented, and easy to audit. Python with libraries such as spaCy, Hugging Face Transformers, pandas, and scikit-learn—or an equivalent stack—makes the most sense to me, but I’m open to a better suggestion as long as it can be containerised for deployment. Acceptance criteria (all must be met): • A preprocessing script that ingests my sample data folder and outputs a tidy feature set • A trained model that beats the baseline I will provide on a held-out test set • A Jupyter notebook (or Markdown report) explaining methodology, key metrics, and error analysis • Dockerfile or conda environment file that recreates the exact runtime Once we have these pieces in place, I’ll integrate the model into my internal dashboard and request a brief hand-off call to cover maintenance pointers.