Callum Curtis

Developed data privacy vaults. • Expanded support for Skyflow's data vault product by delivering a BigQuery integration using Go, Gin, GCP Cloud Run, and Terraform for a Fortune 10 customer. • Built an evaluation framework for named entity recognition (NER) models responsible for redacting personally identifiable information (PII) in unstructured data using seqeval, nervaluate, Hugging Face Tokenizers, and spaCy. • Instrumented cryptographic operations across 4+ microservices using Go, Prometheus, and Grafana.

Built recommendation systems for the Shop app. • Implemented a two-tower neural network to enhance home feed personalization for the Shop app using PyTorch, TorchRec, BigQuery, Airflow, and Kubernetes. • Designed complementary-product models to improve recommendations for cold-start users using co-purchase data and BigQuery. • Trained, tuned, and evaluated an ALS recommendation model using Spark and MLlib. • Created a pipeline producing product embeddings daily for LLM recommendation models. • Personally ideated and prototyped a tool for comparing internal recommendation systems using Streamlit over a weekend, driving company-wide adoption and executive-level visibility, including demos to the CEO and CTO.

Implemented machine learning models and managed data to detect threats within customers’ digital environments. • Developed and scaled unsupervised machine learning models for detecting insider threats across customer networks and devices through behavioral analysis using Spark and Scala. • Constructed a fine-grained monitoring service for data pipeline and model serving costs. • Built an API for LLMs to interact with threat detections using FastAPI, increasing monetization on the Azure Marketplace by enabling integration with Microsoft Copilot for Security. • Established CI workflows and developer tools to validate data pipelines.

Contributed to web applications for real-time collection, analysis, and visualization of data from sensors aboard recreational, industrial, and government ships. • Developed full-stack features for fleets of sensor hubs sold by 100+ retailers, allowing end-users to remotely monitor their property from anywhere in the world through real-time video and sensor readings. • Architected a unified framework for implementing persistence, analytics, and REST APIs for sensor integrations using Express, Node.js, TypeScript, and SQLite. • Constructed frontend views for sensor configuration and data monitoring using React and TypeScript, achieving performant video playback and dashboards with 80k+ data points hosted by resource-constrained edge devices. • Implemented support for an additional third-party sensor type using Python.

Constructed data workloads analyzing sessions from embedded devices. • Expanded pipelines for processing data from 110k+ hours of device telemetry, generating high-level reports for team leads and enriched views for engineers and quality assurance. • Wrote a simulator for embedded software, reducing resolution time for a release-blocking bug by 70%+. • Traced performance degradation in a computer cluster back to BLAS libraries used by third-party Python dependencies, addressing the root cause and reducing memory usage by 30% • Received 100% performance ratings, including commitment to quality and ability to handle ambiguity.