§ Designing a reinforcement-learning-based portfolio optimization algorithm for a tier 1 US bank with Stable Baselines3 and Gym, deployed on GCP and served through an in-house API
§ Developing a novel supervised-learning method that mimics RL, greatly reducing training time by directly optimizing against Sharpe as an objective function; achieved competitive out-of-sample investment metrics
§ Collaborating with Banco de Crédito BCP's FX trading desk to assess efficacy of RL agents in trading a target currency pair, leveraging the models' ability to capture complex market dynamics

§ Engineered end-to-end ETL pipelines with Dagster to consolidate multi-source data from internal CX services, ensuring hourly integration with the Snowflake warehouse
§ Developed dbt staging and intermediate models to transform raw data to production-quality datasets and compute aggregate KPIs, orchestrated with CI/CD pipelines via GitHub Actions
§ Built dynamic Metabase dashboards with Snowflake SQL, enabling 220+ users to monitor CX team performance and productivity
§ Trained and deployed LSTM models in PyTorch for time series forecasting of customer service call volumes, achieving low out-of-sample RMSE and optimizing future agent scheduling

Python, Dagster, EC2, S3, Flask, JS, fetch
§ Built BetterBet, a sports betting web app displaying arbitrage and low-hold opportunities
§ Integrated Dagster for asset-centric orchestration, ETL, and performing asset checks and validations; pulling betting odds from OddsBlaze API
§ Deployed pipeline on EC2, using cron to trigger jobs every minute; pushed processed data to S3
§ Hosted a Flask API on EC2 with RESTful endpoints to dynamically retrieve data from S3; implemented frontend fetch requests for live user interaction