A collection of open-source tools, libraries, and experiments I've built.
Building agentic workflows in Django shouldn't require a pile of custom glue. This package wraps the OpenAI Agents SDK with Django-native sessions, runs, and background tasks, so you can kick off multi-step agents and poll status from any UI.
Feeding random text to an LLM is easy; building a reliable knowledge base is hard. This tool automates the creation of structured, domain-specific knowledge "compendiums" to ground your AI applications in reality.
A composable, deterministic-friendly text-generation toolkit. I ported (and expanded) my original Swift `Wordsmith` library to Python because I needed reliable, seeded procedural name and title generation for other projects.
If you're simulating NPCs or fictional characters, you need them to feel distinct. This library generates full Big Five personality profiles—complete with life-stage biases and conflict-resolution styles—for storytelling and testing.
Colors are hard. This library makes them easy. It provides a Pythonic API for HSBA colors and includes factory methods for generating harmonious color schemes like split complementary, triadic, and monochromatic palettes.
Writing complex prompts shouldn't feel like guessing. I built this to treat prompts like code—structured, modular, and written in markdown, so you can compose and test them with the same rigor as your Python logic.
Stop re-running your expensive data processing steps every time you change one line of code. This tool intelligently caches intermediate steps in your pipeline, letting you iterate on the end of your workflow instantly.
Ever tried to get an LLM to read its own API docs? It's a mess. This CLI tool grabs the massive OpenAI spec and compiles it into clean, deterministic, AI-friendly HTML blocks that your agents can actually understand and use.
Releases should not start with a blank page, so I built this GitHub Action to draft release notes from tag diffs using the OpenAI Responses API. It pulls commit messages and focused diff slices into a clean draft you can ship fast.
Need to test an HR system? Generating believable test data is a pain. `fakedin` uses AI and Faker to hallucinate shockingly realistic résumés and job descriptions that look and feel like the real thing.
An experiment in generative storytelling. It takes an image and uses multimodal LLMs to weave a dynamic, unfolding narrative around it, creating a feedback loop between visual and textual creativity.
A complete AI recruiting workspace I built with Django and Celery. It used explainable AI to match resumes to job descriptions before that was cool.
A specialized Django application for translating Gurmukhi text to English, focused on preserving the nuances of the original script.
Sometimes you just need to feed your entire codebase into ChatGPT. Coderoller smashes your whole source tree into a single, organized Markdown file so you can drag-and-drop your context in one go.
Before OpenAI had an official Swift library, I built this. It offered a clean, Swifty interface for chat completions, embeddings, and token management.
An experimental system for generating procedural music based on "style genotypes," allowing you to breed songs like digital plants.
A procedural generation engine for creating 2D sci-fi spaceship sprites. Useful for populating game worlds with infinite unique vessels.
Generates those little mechanical details (greebles) that make sci-fi surfaces look complex. A fun exercise in recursive 2D geometry.
A Swift library for modeling and generating aesthetic concepts like color palettes and schemes, making generative art more harmonious.
A tool for procedural text generation—names, places, items. Perfect for when you need a thousand unique names for your RPG NPCs.
A utility belt for Swift Strings. It adds all those missing convenience methods you wish the standard library had for text manipulation.
Simulates human-like personalities using the Big Five (OCEAN) model. Great for giving game characters distinct behavioral traits.
RNG is boring. This library provides sophisticated randomness tools, including Gaussian distributions and probability bags, for more organic procedural generation.