Voice AI Pipelines
Building production voice agents end-to-end: speech-to-text, LLM reasoning, and text-to-speech. I focus on the details that decide whether users stay, such as latency, conversation quality, and multilingual performance.
Hello, I'm
Noah Darwich
ML Engineer & Research Data Scientist
I'm an ML Engineer working on the voice pipeline — STT, LLM, TTS — for voice agents in African and Arabic-speaking markets. Before that, six years of research data science: building multilingual datasets and shipping LLM tools for academic teams at Yale, Oxford, Georgetown, Princeton, and the Rockwool Foundation.
What I Do
Multilingual & Arabic NLP
Native Arabic speaker working on models and data for African and Arabic-speaking markets. Six years of building and auditing multilingual datasets means I know where language models break, and how to measure it.
ML Systems & Data Engineering
From raw, messy data to deployed models: dataset construction, quality control, LLM-powered extraction tools, and full-stack ML services like PRO-TEST, an ensemble model with a REST API, CI/CD, and monitoring.
AI/LLM-Powered Research Automation
Develop custom, Python-based AI and computational tools to automate and streamline multi-step academic research workflows. This process includes leveraging Large Language Models (LLM) and APIs to efficiently extract, categorize, and structure complex variables, significantly reducing manual effort in coding and data preparation.
Things I've built
Automate Research Data Collection with LLM Intelligence
Extract entities, structure events, and build quantitative datasets from qualitative sources — designed for researchers.
PRO-TEST v2.0: Protest Outcome Prediction
Ensemble model trained on 13,000+ protest events, served via a REST API with a live frontend.
Institutions I've worked with
