Hi, I'm Sam O'Nuallain

Full-stack AI/ML engineer building production systems and publishing research. Second author on EMNLP '25 paper, deployed MCP servers at Klaviyo, achieved SOTA on BRIGHT benchmark, and scaled LLM training pipelines 15x with parallel orchestration.

Get in touch View projects

MCP servers + RAG systems

Published research (EMNLP)

GPU training at scale (A100)

React + Python full-stack

Ambient visual loading…

Klaviyo • Internship

Software Engineering Intern — shipped analytics for 400+ senders

Owned end-to-end lifecycle of a Model Context Protocol (MCP) server extension with threaded orchestration layer for complex data retrievals.

Shipped push notification analytics (Python, React) adopted by 400+ companies, generating 5,000+ reports in two weeks.

PythonReactAWSAPI analytics

Product adoption

400+ companies

using the push analytics feature

Report throughput

5,000 reports / 2 wks

customer-facing dashboards

Roadmap velocity

3-week ship

MCP server feature lifecycle

Vector search service

CLIP-powered similarity with FastAPI

Built a pipeline to embed and serve image vectors for high-throughput search.

Developed an adaptive batching with a queue-and-timeout mechanism to maximize GPU utilization.

PythonFastAPICLIP

Dataset scale

100k images embedded

Spark + Kafka pipeline

Inference speed

162 QPS (8-core)

FastAPI CLIP service

Deployment

Dockerized end-to-end

X-Camp Academy • Remote

Prototype chatbot + coding tutor agent

Implemented cookie-based session management for a prototype chatbot, maintaining persistent state and conversation context across sessions (Flask, React).

Automated conversion of 90,000+ student submissions into an embedding index and integrated it into a FAISS-backed dense retrieval system, improving LLM-judged response quality by 42%.

FlaskReactFAISSEmbeddings

Embedding index

90k+ submissions

automated conversion pipeline

Response quality

+42% improvement

LLM-judged evaluation

Session state

Cookie-based

persistent context management

AI Poker Agent with Deep Reinforcement Learning

Monte Carlo planning that learns

Engineered a deep reinforcement learning agent for Texas Hold'em, leveraging a Monte Carlo Tree Search (MCTS) guided by a custom deep learning policy network to learn decision-making under uncertainty; secured 7th place in a 22-team tournament.

Policy updates stay stable by mixing search-backed moves with replay buffers.

PythonPyTorchMCTS

Simulations run

50,000+ poker games

self-play training

Policy design

AlphaZero-inspired

custom network for Hold'em

EMNLP '25 Publication • UMass CIIR

Second author — RaDeR retrieval paper

Research paper on enhancing dense retrieval models through retrieval augmentation techniques for improved information retrieval performance.

Fine-tuned novel retrieval models on A100 clusters, achieving state-of-the-art results on BRIGHT benchmark with DeepSpeed optimization.

PythonHuggingFacePyTorch

Publication

EMNLP '25

second author, SOTA on BRIGHT

Data pipeline

15x throughput boost

vLLM GPU data generation

Systems & skills

A stack tuned for shipping research

End-to-end delivery from modeling to interfaces: lean services, ML pipelines, and the tooling that keeps experiments reproducible and shippable.

PythonBackend • ML

JavaScriptFrontend services

C++Performance systems

PyTorchModeling & research

SQLAnalytics & storage

React.jsInterfaces

DockerContainers

GIT

GitCollaboration

FastAPIAPIs at speed

Django RESTStructured services

NLP / HuggingFaceToken & embedding stacks

LangChainLLM orchestration

Education

University of Massachusetts Amherst

MS in Computer Science

Expected May 2026

University of Massachusetts Amherst

BS in Computer Science & Economics

Graduated December 2024

•

3.9 GPA

Publications

RaDeR: Reasoning-aware Dense Retrieval Models

D. Das, S. O'Nuallain, R. Rahimi

EMNLP '25