Projects
Most of the projects listed here are open source, reflecting how I like to work: exploring ideas in public and sharing what I learn with people interested in similar problems. Over the years, that has taken me through data publishing and orchestration, data science on the web, networking, interfaces, browser and IDE extensions, system programming, and machine learning systems.
Agentic AI
An agentic writing platform where AI agents learn an organization's writing style from docs and links, then draft, review, and refine documents in the right voice. Built around style inference, multi-agent collaboration, and human-in-the-loop editing.
A meeting intelligence system that turns recordings into searchable transcripts, decisions, action items, and highlight clips, then feeds that context into downstream strategy workflows. Built for automated ingestion, summarization, speaker attribution, and knowledge retrieval.
An agentic workflow for sourcing, ranking, and triaging job listings into a single dashboard. Built to automate data collection, summarize roles, prioritize opportunities based on user intent, and support the full job search process with agents that help answer technical questions, rewrite multiple CVs for different roles, manage applications, and prepare for interviews.
JavaScript · Data Science · ML on the Web
Open-source JavaScript library providing high-performance, Pandas-like data structures for manipulating and processing structured data in the browser and Node.js.
Co-created the project and contributed to the core dataframe data structure and it operations, plus the general architecture and design of the library
Co-created the project and contributed to the core dataframe data structure and it operations, plus the general architecture and design of the library
An interactive JavaScript notebook environment - think Jupyter but for the browser.
I impemented the core notebook data model, cell execution logic etc.
I impemented the core notebook data model, cell execution logic etc.
A TypeScript port of scikit-learn for JS environments (browser, Node.js), powered by TensorFlow.js.
Contributed to project planning, the initial kickstart, and model saving implementation.
Contributed to project planning, the initial kickstart, and model saving implementation.
Machine learning and data science library for JavaScript and TypeScript, powered by TensorFlow.js.
Contributed feature engineering utilities and preprocessing pipelines — a building block for training and deploying ML models entirely in-browser or on Node.
Contributed feature engineering utilities and preprocessing pipelines — a building block for training and deploying ML models entirely in-browser or on Node.
Open Data · Publishing Systems
The world's leading open-source data portal platform — used by governments and organisations globally to publish and share datasets.
Contributed to adding new features, python 2 to 3 migration, bug fixes, documentation and creating varieties of extensions to improve data publishing workflows.
Contributed to adding new features, python 2 to 3 migration, bug fixes, documentation and creating varieties of extensions to improve data publishing workflows.
A modern JavaScript/React framework built on Next.js for rapidly building rich, feature-complete open data portals and publishing systems. Natively supports CKAN, GitHub, Frictionless Data Packages, and more.
Helped with the initial project kickstart and idea consolidation, and contributed to integrating the framework into various data portals.
Helped with the initial project kickstart and idea consolidation, and contributed to integrating the framework into various data portals.
Systems · Infrastructure · Machine Learning
A Postgres extension written in Rust that hooks into the executor and replication pipeline to track updates on a chosen table column and mirror those values into Redis in real time. A deep dive into Postgres internals — shared memory, hooks, and the extension API.
A minimal VM-based job runner that uses Firecracker micro-VMs to securely isolate and execute arbitrary user scripts. Each job gets its own ephemeral Linux VM with a mounted filesystem. Covers networking, Linux kernel images, root filesystems, and the Firecracker API end-to-end.
Lightweight Go runtime for deploying and serving scikit-learn models exported as ONNX. Implements the ONNX graph execution engine from scratch in Go. Built for production-grade ML inference without a Python runtime dependency.
Interfaces · Developer Tools
