Jeffrey Lee Conradi

Developed production-grade Python software and AI solutions supporting property operations, visual inspection, asset monitoring, occupancy analytics, and operational automation. Designed backend services and APIs using Python (FastAPI/Flask), REST and gRPC with asynchronous programming to integrate production ML workflows. Built scalable service components for real-time ML inference, created data processing/modeling pipelines using Spark/PySpark, Databricks, Airflow, AWS Glue, and S3, and deployed across AWS (EC2, ECS, S3, Lambda) with Docker, Kubernetes, and Ubuntu. Improved dataset preparation via automated annotation/segmentation workflows (41% reduction in annotation effort) and enhanced monitoring with CloudWatch, structured logs, inference latency and GPU metrics, drift detection, and automated alerting. Optimized performance to reduce p95 inference latency by 39% through batching, async processing, memory management, and profiling. Implemented secure service workflows using OAu

Architected a real-time AI risk detection system integrating 200+ CCTV streams on Google Cloud (GKE, Cloud Run, Pub/Sub), reducing incident detection time by 40%. Designed scalable ML pipelines for computer vision and NLP workloads, improving processing throughput by 35%. Built a CV+NLP document intelligence platform improving automated classification accuracy by 30% and fine-tuned LLM-based recommendation models using LangChain and OpenAI APIs to improve predictive relevance by 30%. Implemented model versioning/lifecycle management for more reproducible deployments (35% improvement), achieved 85%+ test coverage with Pytest and Cypress, and delivered a secure multi-tenant cloud architecture meeting enterprise privacy and compliance requirements.

Developed an NLP-based recommendation engine using CRM and customer interaction data to improve sentiment prediction accuracy by 28%. Built an LDA-based document classification pipeline reducing manual case tagging effort by 40%, and implemented clustering/segmentation models improving sales targeting efficiency by 25%. Optimized distributed Spark ML pipelines reducing model training time by 30% and engineered scalable model inference services to improve response latency by 20%.