Machine Learning Predictive Analytics Developer
Budget: ₹37,500 – ₹75,000 INR
The project involves building a complete predictive-analytics engine that ingests structured data from our existing spreadsheets and SQL databases, then delivers clear, actionable forecasts inside a user-friendly application.
Core tasks
• Design data-processing pipelines: cleaning, feature engineering and pattern recognition on millions of structured rows.
• Develop and train machine-learning and deep-learning models—scikit-learn for baselines, then TensorFlow / PyTorch for advanced networks—focused on maximising predictive accuracy.
• Validate results with cross-validation and well-chosen metrics (RMSE, AUC, etc.) and iterate until agreed performance thresholds are reached.
• Wrap the final model in a lightweight API and integrate it into a desktop or web front-end (Python + FastAPI or similar).
Deliverables
1. Reproducible codebase with clear comments and requirements.txt.
2. Trained model artefacts and versioned datasets.
3. Deployment-ready application with step-by-step run instructions.
4. Brief report explaining model choices, feature importance and how to retrain with new data.
Acceptance criteria
• End-to-end pipeline runs on my machine in one command.
• Predictive performance meets or exceeds the benchmark we set together.
• Code passes a quick peer review for clarity and PEP 8 compliance.
While the immediate focus is predictive analysis on structured data, the design should stay flexible so we can later experiment with unstructured inputs or move toward prescriptive insights without rewriting the entire stack.
Core tasks
• Design data-processing pipelines: cleaning, feature engineering and pattern recognition on millions of structured rows.
• Develop and train machine-learning and deep-learning models—scikit-learn for baselines, then TensorFlow / PyTorch for advanced networks—focused on maximising predictive accuracy.
• Validate results with cross-validation and well-chosen metrics (RMSE, AUC, etc.) and iterate until agreed performance thresholds are reached.
• Wrap the final model in a lightweight API and integrate it into a desktop or web front-end (Python + FastAPI or similar).
Deliverables
1. Reproducible codebase with clear comments and requirements.txt.
2. Trained model artefacts and versioned datasets.
3. Deployment-ready application with step-by-step run instructions.
4. Brief report explaining model choices, feature importance and how to retrain with new data.
Acceptance criteria
• End-to-end pipeline runs on my machine in one command.
• Predictive performance meets or exceeds the benchmark we set together.
• Code passes a quick peer review for clarity and PEP 8 compliance.
While the immediate focus is predictive analysis on structured data, the design should stay flexible so we can later experiment with unstructured inputs or move toward prescriptive insights without rewriting the entire stack.