Looking for a job

Nathan Behailu

@nathanbehailu

Message

I build full-stack AI products and machine learning systems from ambiguous problems.

United Arab Emirates

Message

At Caisson Ventures, I build MVP digital products for dental claims and HVAC service automation, turning ambiguous requirements into testable workflows. My full-stack prototypes use Next.js, FastAPI, and Supabase to support rapid validation and client-facing demos.

I've also built research and data systems at NYU, including a pipeline that transformed 40,000+ scanned documents into 2M+ structured records. Across AI projects, I develop fair and reliable ML evaluation frameworks, predictive models, and deployed applications.

Experience

Work history, roles, and key accomplishments

Current

Builder (Software Engineer)

Current

Caisson Ventures

Sep 2025 - Present (1 year)

Defined and delivered MVP digital products for dental claims and HVAC service automation. Built full-stack prototypes using Next.js, FastAPI, and Supabase, and developed rules-based workflow engines to automate documentation and dispatch logic.

Next.js fastAPI Supabase Product Development Workflow Automation

Federated Learning Research Assistant

New York University, MOMA Lab

May 2025 - Jul 2025 (2 months)

Conducted applied research on fairness and trust trade-offs in distributed AI systems. Designed client-side evaluation frameworks to test system behavior across federated rounds.

Federated Learning Fairness Distributed Systems Evaluation Frameworks

Big Data Research Assistant

New York University

Jan 2024 - May 2025 (1 year 4 months)

Built and maintained an end-to-end data pipeline converting 40,000+ scanned documents into 2M+ structured records. Improved data extraction accuracy by 70% and implemented retrieval-augmented generation techniques to reduce manual review overhead.

Data Pipelines Data Extraction RAG Python

Machine Learning Research Assistant

Center for Responsible AI at NYU

Sep 2024 - Feb 2025 (5 months)

Developed optimization frameworks balancing model performance and fairness. Applied post-processing techniques to reduce demographic bias and built reproducible evaluation pipelines using Scikit-learn and Fairlearn.

scikit learn Fairlearn Model Optimization Fairness Evaluation Pipelines