New York, New York, United States
As a kid, I didn’t just play with machines; I wondered what it would take to make them think. Today, that childhood curiosity has evolved into a concrete engineering reality: bridging the gap between high-scale data engineering, mathematical optimization, and the physical world. The biggest challenge in modern industry isn't a lack of data; it's that data is often trapped on a screen, entirely disconnected from physical execution. A cloud database can process millions of transactions in milliseconds, but a warehouse forklift still moves at exactly five miles per hour. If a high-velocity pricing algorithm or a supply chain pipeline isn't perfectly synchronized with actual physical space and inventory, the system breaks. That specific friction point where massive, multi-million-record data infrastructure collides with physical machinery is where I build. With four years of experience spanning data science, scalable cloud engineering, and Physical AI, I design end-to-end autonomous systems that don't just analyze the world but actively sense, move, and interact with it in real time. Core Pillars of Execution: • Operations Research & ML: Translating complex business bottlenecks into deterministic mathematical execution. Expert in Mixed-Integer Linear Programming (MILP), demand forecasting, and dynamic pricing velocity models. • High-Scale Data Engineering: Designing the real-time, low-latency ETL/ELT data pipelines and cloud infrastructure (GCP, Cloud Run, Docker) required to keep models running with 99.9% production reliability. • Physical AI & Computer Vision: Bringing AI off the screen. Applying real-time computer vision (YOLOv8), sensor fusion, and spatial reasoning to automate physical environments, warehouse logistics, and robotic workflows. I don't build AI that lives in a vacuum. From migrating massive backend architectures without a second of downtime to designing vision systems for physical automation, I build scalable, explainable solutions that turn chaotic operational constraints into high-margin revenue drivers. Let's push the boundaries of what intelligence can do when it finally comes off the screen and into the concrete world.
Focus: Python Pipelines, GCP (Cloud Run, Cloud Scheduler), Docker, MongoDB, Agentic AI • Architected end-to-end Python ETL/ELT pipelines (pandas, pymongo) syncing a 1.7M+ record MongoDB inventory database across 20+ collections with external vendor APIs (Ingram Book Company), maintaining 99.9% data-reliability SLA. • Containerized Python micro-services using Docker and deployed to GCP Cloud Run, leveraging Cloud Run Jobs and Cloud Scheduler to orchestrate multi-tier refresh cadences: 4-hour micro-batches for stock flow, daily syncs for out-of-print items, and weekly bulk processing for the full catalogue. • Implemented a CI/CD workflow (GitHub Actions) to automate Docker image builds and Cloud Run deployments, reducing manual release effort and enabling same-day hot-fixes for production pipeline issues. • Engineered competition-based dynamic pricing models in Python by integrating real-time demand signals and automated competitor market research feeds, contributing directly to marketplace volume and margin optimization. • Designed data architecture for a Warehouse Management System and performed continuous root-cause analysis on CDF/EDI ingestion pipelines (FileZilla server), sustaining 99.9% data-transmission reliability. • Led the Agentic Commerce project, integrating conversational LLMs with Stripe and Google Merchant APIs to orchestrate autonomous chat-to-checkout workflows, reducing manual pipeline steps by ~40%. • Provisioned a GCP Cloud Run service end-to-end using Terraform (GCP provider): authored main.tf, variables.tf, and outputs.tf; executed terraform init / plan / apply to deploy a containerized Python service — published the working configuration on GitHub as a reusable IaC reference alongside existing google cloud-based workflows.
• Co-authored a research paper on “How to create a real-time Robot Mental Model on UE 5 using ADAPT cognitive architecture.” • Rendered the real world to a virtual world in real-time using camera, LIDAR, and other sensors for AI implementations. • Imported and utilized actual robot URDF, FBX, OBJ files within the simulation environment using Blender & MeshLab. • Achieved 95% accuracy in environmental mapping and employed ISSAC Simulation and Unreal Engine for creating simulation. • Integrated ROS, simulation tools (Gazebo), LLMs (OpenAI), and GenAI framework (LangChain) for robots to execute written commands (e.g. “get out of the tunnel”, “explore the map”), with an 80% success rate in command execution. • Integrated YOLOv8 with ZED stereo camera to estimate object dimensions using 5×5 point cloud patch averaging with accuracy. • Tested on 30+ samples (phones) at 350–400 mm depth, achieving <3% error with active depth sensing and object measurement. • Developed object detection models for robots using the YOLO algorithm (CNN), detecting 2D/3D objects in images and videos. • Leveraged PyTorch and TensorFlow to develop autonomous decision-making for robots, achieving 95% accuracy.
• Artificial Intelligence (CS 627/385), Data Mining (CS 619), Linux/Unix (CS 648) • Evaluated AI, Data Mining, and Linux/Unix courses, providing feedback to 300+ students, and enhancing their performances. • Facilitated doubt sessions, workshops, provided course materials, answered questions, and mentored students. • Maintained accurate records of grades and attendance, assisted in exam administration and proctoring.
Focus: ETL Pipelines, Azure Cloud, Machine Learning • Developed and deployed Machine Learning models (XGBoost, Random Forest, RNN) on 10 years of historical HR data to predict employee churn risk and forecast future resource availability across fiscal quarters. • Enabled retention strategies by identifying high-risk employees with 85% accuracy prior to appraisal cycles. • Built end-to-end data engineering pipelines using Azure Data Factory (ADF) and SQL to ingest real-time metrics (bench status, skill acquisition, work hours). • Orchestrated Azure Data Factory (ADF) pipelines and Azure Databricks workflows to ingest and transform data from 8 source systems, reducing data latency from 4 hours to under 30 minutes to power near-real-time executive dashboards. • Configured event-based triggers to automatically retrain models and refresh dashboards, ensuring up-to-date operational intelligence. • Implemented NLP-based sentiment analysis on employee feedback to correlate "work enthusiasm" with project delivery risks. • Provided actionable insights for resource allocation and promotion planning, directly supporting long-term workforce stability.