Rushikesh
Baban Kedar

Structures messy financial histories and unmapped transactional records using an end-to-end XGBoost machine learning model pipeline. Integrates an asynchronous computer vision segmentation layer to automate extraction details straight from raw transactional media feeds.

Powered by 9 concurrent asynchronous AI agents working together to query and interpret unstructured financial data streams in real time. Deploys an integrated semantic search loop over FAISS vector spaces.

End-to-end ML classification pipeline predicting membership propensity upgrades over unbalanced features. Applied SMOTE balancing and benchmarked XGBoost arrays to hit an AUC-ROC of 0.7969.

An advanced multi-turn conversation engine structured entirely around a LangGraph relational graph schema. Employs deep state memory checkpoint parameters to drive text drift errors down by 90%.

Aggregates postings from 7 distinct scrapers, sorting matches based on calculated TF-IDF text similarity maps, and pushing structured results over background Telegram and email worker pipelines.

Transformer-based computer vision and text parsing pipeline. Vision Transformer (ViT) encoders isolate target image matrices while a GPT-2 decoder maps fluid semantic context outputs seamlessly.

End-to-end data analytics pipeline predicting PM2.5 concentrations across 5 global cities. Features extensive EDA, time-series forecasting, and advanced XGBoost regression modeling.

Empirical examination of monetary policy transmission into CPI inflation. Utilized statistical testing to identify a 5-year policy transmission delay from the RBI Repo Rate.

High-fidelity BI dashboard ingesting retail datasets to deliver spatial-temporal insights, profitability tracking, and dynamic customer behavior drill-downs for stakeholders.

Fully automated dashboard consolidating cross-geography databases. Engineered a macro-free architecture for tracking real-time attrition, risk management, and compliance alerts.