Ehsan ML Engineer · Applied Data Scientist

Building Machine Learning Systems & Decision Intelligence

I design data pipelines, machine learning systems, and AI agents that automate real-world workflows.

Selected Works

● Completed

DataForge Bench — ML Data Format Benchmarking Tool

Real-time benchmarking system for comparing JSON, CSV, Parquet, Protobuf, and Avro across serialization speed, storage efficiency, streaming throughput, feature engineering, and ML training — fully client-side with optional Python backend for high-fidelity evaluation.

React · Vite · JavaScript · PapaParse · Python · PyArrow · FastAvro · Protobuf · NumPy · Pandas
🚀 Live Dashboard
● Completed

Uplift (Decision Impact) Modeling MLOps Pipeline

Production-grade machine learning system for causal uplift modeling, deployed with FastAPI, Docker, and Kubernetes, with CI/CD automation and real-time monitoring using Prometheus and Grafana.

Transit Dashboard Preview

API · Kubernetes · Prometheus · Grafana

Python · FastAPI · LightGBM · GTFS-RT Protobuf · NetworkX · React · Docker
● Completed

Causal Marketing Optimization & Budget Allocation System

End-to-end machine learning system for uplift modeling and budget optimization using 45M+ transaction records. Instead of predicting conversions, the system estimates individual treatment effects and determines the optimal targeting strategy under budget constraints, improving campaign ROI by over 60% compared to random targeting.

🚀 Live Dashboard
Budget Optimization Dashboard

Uplift Modeling · Budget Optimization · streamlit · Monitoring

Python · LightGBM · Scikit-learn · Pandas · Streamlit · FastAPI · Docker · streamlit
● Completed

Real-Time Transit Delay Prediction System

End-to-end machine learning system that ingests real-time transit data, performs spatiotemporal feature engineering, and predicts delays (5–30 min horizon) via a low-latency API, with integrated monitoring and live dashboard.

Transit Dashboard Preview
Python · FastAPI · LightGBM · GTFS-RT Protobuf · NetworkX · React · Docker
● In Progress

Real-Time Credit Risk & Approval Intelligence System

Production-grade decision intelligence system that goes beyond default prediction to optimize financial decisions under risk constraints. Combines LightGBM risk scoring with constrained portfolio optimization (maximize profit subject to budget and default rate limits), an AI policy agent powered by Groq LLM, and real-time model drift monitoring — all deployed as a full-stack Docker application with FastAPI, Next.js, and PostgreSQL.

ML Risk Scoring · Portfolio Optimization · AI Agent · Drift Monitoring · Sentry

Python · FastAPI · LightGBM · SHAP · Groq LLM · PostgreSQL · pgvector · Next.js · Docker · OpenTofu · GitHub Actions · Sentry
● In Progress

AI Outreach Agent

Agent that generates emails, enriches leads, and automates scheduling.

OpenAI · LangChain · APIs
● In Progress

Data Enrichment Pipeline

Scraping + enrichment + validation pipeline for structured data.

Python · Scraping · APIs

Skills

Data Engineering: ETL, Airflow, Kafka, SQL
Data Science: ML, Feature Engineering, Modeling
AI: LLMs, LangChain, Automation Agents

Contact

Email | GitHub | LinkedIn