At ThirtySignals, I created DSPy prompt optimization workflows that improved resume extraction accuracy from 56.8% to 83.2%. I also trained LightGBM forecasting models on ad, adset, and campaign records.
I productionized Dagster forecasting pipelines, reducing processing time by 60%, and integrated MLflow tracking and model registry workflows with an LLM decision layer. I also built core workflows for Pupil, an open-source AI agent evaluation framework.
At Infosys Springboard, I implemented a real-time object recognition system using YOLO-based models. The custom-dataset models achieved 92% detection accuracy across 80 object categories, and I deployed the pipeline with Python, OpenCV, and Streamlit.
In my projects, I built VisionAssist, a low-light obstacle detection device, and a real-time fall detection system using edge-deployable machine learning. I also built MailMind, an AI-powered Gmail assistant for inbox sync, triage, hybrid search, and safe cleanup review.

