Reza Hosseini

• Architected an agent-operated system that turns each podcast episode into 100 publish-ready Instagram Reels across five editing styles, cutting days of manual editing to a ~63-minute GPU render. • Designed a four-stage LLM-driven workflow with schema-validated JSON edit plans, multimodal speaker attribution and word-level caption alignment (faster-whisper). • Optimized GPU rendering with NVENC and CuPy (98 to 63 min) and embedded QA gates and an A/B style experiment: 400 videos, 100% QA pass rate.

• Built MEWISO, a geospatial AI agent that generates analytical maps and reports from geospatial data and imagery, using LangGraph to orchestrate multi-step spatial analysis and workflow automation, replacing a manual map-and-report process and reducing analysis time. • Built a semantic RAG pipeline for MEWISO using PostgreSQL/ pgvector, combining vector similarity search and geographic metadata filtering to retrieve relevant spatial context, with retrieval and answer quality evaluated using Ragas. • Developed and trained a CNN soil segmentation model on 15K+ satellite images, combining manual and automated annotation workflows in Label Studio to build and refine a large-scale training dataset. • Built an Apache Airflow ETL pipeline to orchestrate preprocessing and feature engineering across ~30GB of satellite imagery, providing structured inputs to the CNN model and MEWISO geospatial agent.

– Designed and implemented T-SQL data models for monitoring water flow velocity, integrating with Power BI dashboards to enable real-time visualization and reporting. – Built and deployed 4+ Power BI dashboards across operations, finance, HR, and marketing, automating reporting workflows, integrating Excel-based data sources, and improving data-driven decision-making.

• Developed a Logistic Regression fraud detection model for a portfolio of 100K+ clients, validated through A/B testing and significantly reducing manual review efforts. • Deployed an XGBoost credit scoring model via FastAPI for real-time inference, coupled with an automated feature engineering pipeline. • Migrated data ingestion from a daily batch process to a real-time streaming pipeline using PySpark and Apache NiFi, enabling live P&L monitoring for clients. • Collaborated on the development of an investment recommendation engine by combining users’ trading behavior with their risk profiles to generate personalized investment recommendations.

– Delivered 5+ BPMS implementations to automate manual business processes using Bizagi and Process-Maker, reducing operational workload and improving process efficiency across multiple workflows. – Led business process re-engineering initiatives using Visual Paradigm, analyzing and redesigning workflows to optimize performance, eliminate bottlenecks, and align processes with business objectives. – Implemented process mining solutions for Procure-to-Pay workflows using Celonis and PM4Py, enabling data-driven analysis of process performance, bottleneck identification, and compliance monitoring. – Developed a custom Python-based process mining library for event log analysis, detecting outliers using process-aware features (Fig4PM).