Work

Duration: Dec 2025 – Current | New York, NY

Performed feature engineering and data preprocessing on trading transaction data using Python and Pandas, applying recursive feature elimination (RFE) to improve model accuracy by 15% and cut feature dimensionality by 30%.
Engineered classification models with Scikit-learn and PyTorch (Random Forest, XGBoost) to predict trade settlement likelihood, hitting an 87% AUC-ROC score and streamlining compliance review for operations teams.
Built an AI-powered client risk assessment system using Python, Pandas, and logistic regression with survival analysis, enabling risk management for 80 high-value investment banking clients.
Developed a predictive analytics pipeline with XGBoost regression to forecast client portfolio risk exposure, improving capital allocation accuracy by 20% and reducing pricing error rate by 12%.
Deployed inference APIs with FastAPI and Docker on AWS EC2, integrating OpenAI and Hugging Face APIs for trade document summarization at 99% uptime with low-latency inference.

Duration: July 2025 – Sep 2025 | New York, NY

Built and shipped core mobile features for a video-first platform enabling video resumes, pitches, and short-form content discovery, using Flutter and Dart across a distributed, production-scale app.
Engineered a real-time Stories feed with live content updates using WebSockets and Provider for state management, improving feed load speed by 22% for 15,000+ active users.
Developed the user profile ("Me") tab and notifications module supporting real-time alerts and interactions, using Flutter, Dart, and Provider across the app's core navigation experience.

Duration: Aug 2022 – Aug 2024 | India

Designed a securities recommendation engine for a capital markets platform using collaborative filtering with SVD in Python and TensorFlow, achieving a 24% improvement in client conversion rates and reducing portfolio churn by 11%.
Developed a real-time algorithmic pricing model with XGBoost and Scikit-learn for a fixed income client, processing 50K+ instruments hourly at sub-100ms inference latency and increasing annual trading revenue by 16%.
Implemented an end-to-end MLOps pipeline using Docker and AWS SageMaker to automate model training, validation, and deployment, cutting the retraining cycle from 8 hours to 45 minutes.
Engineered feature selection and preprocessing workflows on 2M+ client transaction records, applying SHAP-based feature importance analysis to improve credit risk model F1 score by 19% across three production models.
Integrated XGBoost and Random Forest models into production AWS Lambda microservices via FastAPI, serving real-time market risk scoring for 200K+ daily trade requests at sub-150ms latency with zero downtime during peak hours.

made w/ ♥ by Aditya Bhuran — always building something new 🚀 📧