ZazaDragon.xyz

// Portfolio

Stuff I shipped 100% Open Source releases

Built it. Use it. Download it.

01 / Project

Personal Database

Offline-first personal wiki. Organize notes into nested folders, write in markdown, link articles with [[wiki-style syntax]], and search everything instantly with full-text indexing. Built with Tauri and SQLite.

Download for Windows View on GitHub

02 / Project

PLANz

Visual mind map and task planner. Add ideas, drag to prioritize, connect them on a live canvas, and export the whole plan as a numbered prompt or checklist. Portable — runs from a USB stick or as a standalone HTML file.

Download for Windows View on GitHub

                
                
                
                revan-hub — RTX 2060 Super
              

                  $
                  revan-hub
                
───────────────────────────────

                  revan
                  engine connected
                  ✓
                

                  model
                  OLMo-3.1-32B Q4_K_M
                

                  gpu
                  6.2 / 8.0 GB VRAM
                

                  cache
                  3-slot LRU
                  warm
                
───────────────────────────────
 

                  [agent]
                  route_tool
                  →
                  "calculator"
                  0.5s
                

                  [agent]
                  yes_no
                  →
                  "yes"
                  0.4s
                

                  [agent]
                  classify
                  →
                  "positive"
                  0.5s
                

                  [agent]
                  score_5
                  →
                  "8,6,9,7,5"
                  3.0s
                
 

                  throughput
                  128 tok/s
                  prompt |
                  3.3 tok/s
                  gen
                

03 / Project

revan-oss

Stateless decision engine for consumer GPUs. Runs a 32B model and constrains output to 1-15 tokens — turning a slow chatbot into a sub-second decision maker. C++ engine, Rust hub, agents in any language.

View on GitHub

                
                revan-agent — Named Pipe IPC
              
                  $
                  revan-agent.exe
                
───────────────────────────────

                  pipe
                  \\.\pipe\revan-agent
                  ready
                
                  mode
                  lazy — loads on first request
                
                  ipc
                  binary protocol, 28B header
                
───────────────────────────────
 
                  [req]
                  session=42
                  →
                  loaded Hermes-3 8B
                  2.1s
                
                  [req]
                  inference
                  →
                  128 tok @ 42 t/s
                  0.3s
                
                  [kv]
                  snapshot
                  →
                  session 42 saved
                  1.6MB

04 / Project

revan-hub/agent

Local LLM inference over Windows Named Pipes instead of HTTP. Binary protocol with ~0.1ms overhead, persistent multi-turn sessions via KV snapshots, and automatic VRAM management. Single C++ file, any GGUF model.

View on GitHub

                
                yt2 — local video summarizer
              
                  $
                  yt2.py
                  https://youtube.com/watch?v=...
                
                  [1/6]
                  Downloading video
                  ✓
                
                  [2/6]
                  Transcribing (whisper small.en)
                  ✓
                
                  [3/6]
                  Detecting scenes
                  →
                  14 cuts found
                
                  [4/6]
                  Extracting keyframes
                  ✓
                
                  [5/6]
                  Summarizing with mistral
                  ...
                
                  output →
                  Neural Networks Ch1.html
                  (12MB)

05 / Project

YT2

Local YouTube video summarizer. Downloads, transcribes, detects scene changes, and summarizes each section with a local LLM. Outputs a self-contained HTML page with embedded keyframe screenshots. Also generates vertical shorts with word-level captions. Recommended model via Ollama: Kimi-K2.5:cloud — fully free with limits.

View on GitHub

More in the works

Always building something. Check back soon.

// About

About Me

I made a personal database app because I kept losing my notes. Then I made a planning app because I kept losing track of what to build next. See the pattern?

I also run a RuneScape private server called Soulsplit 3, which is either a passion project or a cry for help an advanced exercise in server administration, depending on who you ask.

Everything here is open source. Not because I'm generous — because I know someone will find the bugs faster than I will.

— toti.ba on Discord, if you want to yell at me

GitHub Discord

Rust

Python

C++

TypeScript

Java