Ravishankar Kurma

• Python
• SQL
• Power BI

End-to-end sales analysis for a textiles retailer, from raw transactions to an automated Power BI dashboard.

result+32% data quality−60% manual reporting

• Processed and cleaned 25,000+ transaction records with an ETL pipeline covering deduplication, type enforcement and missing-value handling.
• Wrote SQL to deliver KPIs on revenue, category performance and seasonal demand.
• Automated the Power BI dashboard refresh, cutting manual reporting effort by 60%.

• Python
• Pydantic
• FastAPI
• pytest

Validates untrusted warehouse stock updates from JSON, JSONL and CSV feeds before anything reaches the stock record.

resultcommit · reject · quarantine

• Runs every update through a fixed validation pipeline, so feed content can never change what the agent does.
• Closed Pydantic schemas treat unknown fields as errors, and replay and sequence checks catch duplicate or out-of-order updates.
• Quarantines suspicious items instead of discarding them, so evidence is kept for review.
• Records every decision in a hash-chained audit log, with behavioural tests covering the expected outcomes.

• Python
• scikit-learn
• React
• FastAPI

Groups users by psychometric assessment results and explains each group in plain language.

resultinterpretable user segments

• Segmented users with KMeans clustering on psychometric assessment data.
• Layered LLM-generated explanations over the cluster output to produce reports people could understand.
• Built with a Python and scikit-learn backend, FastAPI and a React front end.

• Python
• scikit-learn
• CNN

A convolutional neural network that classifies MRI scans for signs of Alzheimer's disease.

result90.2% accuracyF1 0.89

• Built and validated a CNN on an MRI dataset, reaching 90.2% accuracy with an F1 score of 0.89.
• Used augmentation, class balancing and cross-validation to control overfitting.