tiny experiments, public drafts, half-built ideas
Built to validate the DHTQueryCoordinator design with measurements instead of assertions — dual-layer trio.CapacityLimiter back-pressure, streamed to a real-time dashboard under load. Archived on purpose: it was a learning project, not a product.
Visual, interactive simulations of security vulnerabilities in libp2p peer-to-peer networks. Master the mechanics of exploitation and defense.
Collection of mini agentic projects based on the AgnoAI framework, when I tried learning AI agents in last year of college
Leetcode Journal is a NextJS Full stack application designed to help developers track, assess, and analyze their problem-solving journey on LeetCode. It provides a better way to show the progress of a developer of their coding journey.
Web app that creates and joins Zoom Video SDK sessions, then layers real-time speech-to-text over them using the Web Speech API — including per-participant tracking and speech-pattern detection for questions, commands, and tone.
Visualize GitHub contributions, allowing users to customize, export, and share their activity as high-quality images on Twitter, and LinkedIn.
A minimal Node.js service that handles the Zoom RTMS handshake and consumes realtime media streams off a live meeting. Written to figure out how the webhook + OAuth credential flow actually fits together before building anything on top of it.
An LLM agent wired to a set of tools for real-time crypto news, social sentiment, token analytics, and web search — built for the Global Agent Hackathon, May 2025.
A from-scratch blockchain implementation — proof of work, chain validation, peer node consensus — bundled together with a client wallet in a single deployable Flask application. Written to understand the mechanics rather than to trust a library for them.
A self-hosted Mastra AI agent backend — tools, workflows, and long-term memory through mem0 — packaged to run as a container instead of a hosted service.
A small Python script that posts to a Discord channel on a user's behalf, driven by a list of messages and their assigned intervals. Exactly as large as it needs to be.
Pulls a video's transcript and turns it into a summary with jump-to timestamps, adjustable length, and multi-language support — so you can decide whether the 40-minute talk is worth the 40 minutes.
A chat interface over the Groq inference API, built to test whether a Python Flask backend could be made to deploy cleanly on Vercel's serverless runtime. It could.
A Python library that renders a trained decision tree as an interactive HTML page instead of a static plot. My first package published to PyPI, which turned out to be the more instructive half of the project.
A command-line to-do tracker built with Python and rich. No server, no sync, no account — a local file and a table that looks good in a terminal.
How can I help?
Or message @0xcrackedDev on X