title: 'Mike Schock' keywords: [artificial intelligence, machine learning, software engineering] ...

• Email: [email protected]
• GitHub: mjschock
• LinkedIn: in/mjschock

• Support the Chief Technology Officer (CTO) in building foundational infrastructure and executing business-critical features
• Contribute to the design and development of software components, architecture decisions, and internal tooling
• Create proof-of-concept (PoC) solutions to validate technical feasibility for upcoming product initiatives
• Provide AI/ML technical expertise and develop proof-of-concept solutions for artificial intelligence and machine learning features and integrations
• Participate in technical planning, regular engineering syncs, and code reviews to ensure high-quality execution and alignment with the product roadmap
• Document key implementation decisions and engineering work for team knowledge sharing

• Worked on ArchiLabs' mission to create the first AI Architect to make all new construction faster & more affordable
• Developed across the full stack, including Revit integrations, React frontend, Supabase, and AI agents powered by LangGraph
• Worked directly with the CEO and CTO founders and the BIM specialist
• Contributed to a fast-moving YC-backed startup focused on solving critical problems in the construction industry
• Improved the AI capabilities to enhance construction planning and design efficiency
• Technologies: React, Supabase, LangGraph, LangChain, LangSmith, Revit integrations, AI agent systems

• Spearheaded orchestration and automation of AI agent training (with each agent an ensemble of PyTorch models) into an MLOps pipeline backed by a self-hosted in-cluster duo of Prefect Server and Agent to run training ad-hoc and on-schedule, with follow-up work demonstrating the migration path from the deprecated Prefect Agent to Kubernetes-native Prefect Worker.
• Rapidly prototyped a working MVP showcasing how we could easily scale the training runs via the Prefect-Ray integration and an in-cluster or Anyscale Cluster, also presenting SkyPilot as a way to abstract Ray and cloud computing resources, optimizing for minimal computational cost or time.
• Modernized the developer experience for the AI Platform team by bringing in Tilt to watch for changes in the Kubernetes manifests for full Docker build/pushes, thereafter updating pods without reload for fast iteration, and providing custom functionality to run data preparation, agent training, and inference pipelines via configurable buttons in the Tilt UI.
• Technologies: Cloud SQL for PostgreSQL, Docker, Google Cloud Platform (GCP), Google Kubernetes Engine (GKE), gRPC, Prefect, Python, PyTorch, Ray, SkyPilot, Tilt

• Served as a Teaching Assistant (TA) for CS 7639: Cyber-Physical Systems Design & Analysis.

• Created a proof of concept (POC) for a data engineering pipeline to extract, transform, and load images and their corresponding labels from various data sources and formats into the COCO dataset format with k-fold train-validation-test splits using the FiftyOne and Albumentations libraries.
• Constructed a POC for a data modeling pipeline to train and validate a PyTorch Faster R-CNN model with various modifications for computer vision tasks such as object detection and image segmentation from a train-validation split output by the data engineering pipeline.
• Assembled a POC for a model deployment pipeline to deploy a model produced by the data modeling pipeline into a local docker container running TorchServe (or SageMaker) to run inference tests upon that model and to trigger the creation of a function that ran on schedule to monitor the deployed model.
• Designed a dashboard using Amazon QuickSite to automatically generate visualizations, including emails pointing to those visualizations, that displayed the performance of the served model and assigned SageMaker GroundTruth jobs for our internal teams to help with data labeling.
• Built a POC framework using Kedro and DVC to join the data engineering, data modeling, and model deployment pipelines, running pipeline components only when artifacts tracked by DVC changed.
• Technologies: Albumentations, Amazon QuickSite, Amazon SageMaker Ground Truth, Docker, DVC, Faster R-CNN, FiftyOne, Kedro, Matplotlib, MobileNet, NumPy, pandas, Python, PyTorch, scikit-learn, TorchServe, torchvision

• Refactored the data ingestion pipeline into more modular components.
• Drove the model explainability implementation.
• Led the development of the forecasting service.
• Built a worker service that operated on graph structures representing machine learning tasks and states.
• Developed various features and addressed bugs in our systems.
• Technologies: Amazon Forecast, Amazon Web Services (AWS), Docker, Docker Compose, JavaScript, LightGBM, Matplotlib, NumPy, pandas, Python, SHAP (SHapley Additive exPlanations), Tableau

• Specialization in Artificial Intelligence