IPL 2026 Predictor
An XGBoost-powered IPL match winner predictor trained on 18 seasons of historical data. Includes a Streamlit app with interactive team selection, impact-player logic, and live win probabilities.
Open-source projects, my Claude Code setup bundle, CLI install guides, and a local-model setup advisor — pick one.
An XGBoost-powered IPL match winner predictor trained on 18 seasons of historical data. Includes a Streamlit app with interactive team selection, impact-player logic, and live win probabilities.
Electron desktop app for Open WebUI — pure black UI, venv installer, and system tray integration. Run your local AI chat interface as a native desktop app without touching the terminal every time.
A local memory system for AI assistants that runs entirely on your own hardware. Store and retrieve context without cloud dependencies — full privacy, zero latency.
A Claude Code skill for generating and publishing blog posts directly to Google Blogger. Automate your blogging workflow with AI. Also available as a Claude skill on Clawhub.
Building more tools and skills in public. Follow on GitHub to stay updated.
Follow on GitHubMy entire Claude Code config in one download. Plugins, skills, rules, hooks, agents, commands — hand the file + prompt to Claude Code, Cursor, or Codex and it sets up everything on your machine.
11
plugins
17
custom skills
29
slash commands
6
sub-agents
How it works
Download setup.md, open the Agent Prompt and copy it, then paste both into Claude Code / Cursor / Codex. The agent backs up your existing ~/.claude/, clones the companion repo, merges files non-destructively, and installs 11 plugins. Takes about a minute.
Install guides for every CLI worth having in your AI workflow. Part of the ongoing TechRex Shorts series.
There is no special "agent integration" for these tools. You install the CLI, authenticate it on your machine once, then just tell your agent what to do. The agent runs the commands in your shell using the credentials the CLI already stored locally.
That's the whole integration. No MCP server needed, no API keys in agent config — the agent just shells out to a CLI that already knows who you are.
Deploy projects, manage env vars, and preview deployments from the terminal.
Create PRs, manage issues, clone repos, and run workflows without leaving your terminal.
Spin up local Supabase, run migrations, and deploy edge functions from your terminal.
Manage AWS services, configure credentials, and automate cloud infrastructure.
Official Google Workspace CLI for managing Drive, Gmail, Calendar, and admin resources from the terminal.
Download models and datasets, push repos, manage auth tokens, and run Hugging Face Hub ops from the terminal.
Tell me your laptop's RAM and GPU. I'll pick a model that fits, show you the exact memory breakdown, and hand you the same llama.cpp launch command I run on my own machine — or the MLX equivalent on a Mac.
Q4_K_M = 4-bit, Q8_0 = 8-bit, UD-IQ2_M = Unsloth Dynamic 2-bit. Lower bits = smaller file + less VRAM, slightly less accuracy. Pick the biggest quant your hardware can fit.-ctk q4_0 -ctv q4_0 compresses it to a quarter of the FP16 size.--n-cpu-moe (MoE only)--n-cpu-moe N tells llama.cpp to keep N expert layers in CPU RAM instead of VRAM. The active experts get streamed to the GPU at run time, so you fit a bigger quant than VRAM alone would allow, with only a small speed cost. The recommended value below is auto-tuned to your VRAM — drop it to 0 for maximum speed if you have headroom, or raise it if you OOM.
Set the base URL on any OpenAI client. Drop in any non-empty key. Send the model alias as model.
http://127.0.0.1:8080/v1 sk-local qwen3.6-moe
⚠️ --host 0.0.0.0 exposes the server on your LAN. Drop it to 127.0.0.1 if you only want local access.
Want walkthroughs for the tools listed here? Watch on the YouTube channel. Want the background on how these projects fit into the TechRex workflow? Read About.