Required stack
필요 기술
PythonPandasScikit-learnFinancial AnalysisStatistical AnalysisData VisualizationPlotlyETLMachine LearningLinear RegressionML Explainability (XAI)StreamlitDashWeb DevelopmentCloud Deployment
Project brief
프로젝트 내용
I am building a decision-support tool that acts like an investment banker for hospital deals. The workflow starts with raw historical statements—revenue, profit, cash flow, debt and related line items—pulled in from spreadsheets or a database. I need this data to be cleaned and validated automatically so that outliers, missing values and inconsistent formats are handled without manual intervention.
Once the dataset is in shape, the system must calculate core ratios—operating and EBITDA margins, liquidity and leverage indicators, year-over-year growth rates, cash conversion, risk flags—and feed them into a machine-learning pipeline. I have no fixed allegiance to any one algorithm, but I would like to begin with a solid linear regression benchmark and keep the door open for more sophisticated models if they outperform.
The model should output a single “investment readiness” score for each hospital, accompanied by clear explanations that highlight drivers such as high debt, robust profitability or rapid growth. These explanations have to be both human-readable and suitable for display alongside the score.
Everything culminates in an interactive dashboard where users can:
• Upload or connect fresh financial files
• Inspect the cleaned dataset and calculated ratios
• View the readiness score with drill-down explanations
• Toggle assumptions, refresh predictions in real time and export a concise report
Deliverables:
1. End-to-end Python (or comparable) codebase covering automated ETL, feature engineering, model training and inference.
2. A lightweight web dashboard—Streamlit, Dash or similar—deployed locally or on a small cloud instance.
3. Brief technical documentation and a walkthrough so new users can reproduce results and retrain the model with updated data.
Acceptance will be based on:
• Successful automation of data cleaning for a sample of multi-year hospital statements.
• Generation of ratios matching standard finance formulas within ±1% tolerance.
• A regression-based model that outputs a score and text explanations in the dashboard.
• Smooth, error-free user flow from data upload to report export.
Once the dataset is in shape, the system must calculate core ratios—operating and EBITDA margins, liquidity and leverage indicators, year-over-year growth rates, cash conversion, risk flags—and feed them into a machine-learning pipeline. I have no fixed allegiance to any one algorithm, but I would like to begin with a solid linear regression benchmark and keep the door open for more sophisticated models if they outperform.
The model should output a single “investment readiness” score for each hospital, accompanied by clear explanations that highlight drivers such as high debt, robust profitability or rapid growth. These explanations have to be both human-readable and suitable for display alongside the score.
Everything culminates in an interactive dashboard where users can:
• Upload or connect fresh financial files
• Inspect the cleaned dataset and calculated ratios
• View the readiness score with drill-down explanations
• Toggle assumptions, refresh predictions in real time and export a concise report
Deliverables:
1. End-to-end Python (or comparable) codebase covering automated ETL, feature engineering, model training and inference.
2. A lightweight web dashboard—Streamlit, Dash or similar—deployed locally or on a small cloud instance.
3. Brief technical documentation and a walkthrough so new users can reproduce results and retrain the model with updated data.
Acceptance will be based on:
• Successful automation of data cleaning for a sample of multi-year hospital statements.
• Generation of ratios matching standard finance formulas within ±1% tolerance.
• A regression-based model that outputs a score and text explanations in the dashboard.
• Smooth, error-free user flow from data upload to report export.