Selected Work
Systems, tools, and product work.
A closer look at engineering decisions across codebase understanding, automation, real-time products, and task execution.
In development
KMS / Ramp
Ramp is a developer onboarding tool designed to help engineers understand unfamiliar codebases. It builds bounded, evidence-backed views of how implementation surfaces relate, then uses those candidates to generate structured onboarding plans. The system is designed to distinguish observed code relationships from assumptions that still require human verification.
Stack
Python, Go, Supabase, LLM integration
Highlights
- —Company-scoped code indexing and evidence extraction
- —Evidence composition across connected implementation surfaces
- —Deterministic candidate selection and onboarding plan generation
- —Persistent onboarding plans and progress
Read case study
Professional experience case study
Blender Automation
An automated 3D asset-to-render workflow that connects asset discovery, AI-assisted requirement extraction, cloud file transfer, and headless Blender rendering. The implementation combines n8n, an LLM integration, Sketchfab assets, Google Drive, a Flask service, and Google Cloud infrastructure. Alongside the working rendering pipeline, the project explored the limitations of generating complex procedural animations through LLM-produced Blender scripts.
Stack
n8n, LLM integration, Sketchfab, Google Drive, Flask, Blender Python, Google Cloud Platform
Highlights
- —Asset requirement extraction and Sketchfab retrieval
- —Asset transfer through Google Drive
- —Headless Blender execution using xvfb-run on a GCP VM
- —Scene construction and PNG or MP4 rendering branches
Read case study
Released project
Guessit
A real-time multiplayer drawing and guessing game built with a React and TypeScript frontend and a Go WebSocket backend. The project combines room-based game coordination, live drawing and chat events, and an operational metrics dashboard. It provided a practical environment for exploring event-driven communication, concurrent server state, WebSocket handling, and observability in a real-time application.
Stack
Go, WebSockets, React, TypeScript, Zustand, Redis
Highlights
- —WebSocket game protocol for room and player lifecycle
- —Drawing and chat events
- —Operational metrics and latency tracking
- —Redis persists metrics; game and room state remain in memory
Read case study
Paused; latest known package version 0.11.2
Nuvom
An open-source Python task execution framework designed around a lightweight, local-first approach to background work. Nuvom provides a task API, queue and result backends, worker execution, retries, scheduling, plugin hooks, and a CLI. Its development explored how to provide useful task-processing capabilities without requiring a Redis-backed infrastructure for the default local workflow.
Highlights
- —@task, .delay(), and .schedule() APIs
- —AST-based task discovery with manifest caching
- —SQLite queue and result backend support
- —Retry and timeout configuration
- —Plugin architecture, CLI, and configuration support
- —One-off, interval, and cron scheduling
- —Automated PyPI release workflow
Read case study