Based in Kathmandu, Nepal. Daily driver is CachyOS / Arch on Hyprland, with quantized models running through llama.cpp. I like machines that stay on my desk — no cloud tax.
I'm Ignaz Bhatta. First-year BSc (Hons) Computer Science with AI student. I have been assembling machines since before I had the words for an OS.
The through-line is simple: take a problem described in words and reduce it to something a machine can answer. Sometimes that answer is a map. Sometimes a rice. Sometimes a model that never leaves the GPU.
Daily-driver config, tabs, and local inference. Host is mine — not a campus billboard.
⡿⠟⠁⠀⠀⠀⠀⠀⠀⠀⠀⢿⣷⣌⠏⣨⣾⡿⠀⠀⠀⠀⠀⠀⠀⠀⠈⠛⠿⢳ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢸⣿⣿⣶⣿⣿⡇⠀⠀⠀⠀⢀⠖⠈⡽⢂⠀⠀⢻ ⡄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠛⠉⠉⠉⠉⠀⠀⠀⠀⠀⠈⡈⣻⡗⠛⠇⡠⣢ ⣿⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠩⠁⠝⠀⣀⠀⠘ ⣿⠇⠀⠀⠀⠀⠀⠀⠀⠀⠀⢀⣀⣀⣀⣀⠀⡀⠀⠀⠀⠀⠀⠀⠀⠀⣻⣿⣧⡄ ⠏⠀⠀⠀⠀⣴⠆⣠⣶⣿⠇⣿⣿⣿⣿⣿⢸⣿⣿⣷⣶⣤⠘⡶⢄⠀⠛⠻⡿⠁ ⠀⠀⠀⢠⣾⣿⢸⣿⣿⠟⣀⣿⣿⣿⣿⣿⡄⠻⣿⣿⣟⢿⣧⢹⣶⣄⠀⠀⠀⠀ ⡦⢀⣴⣿⣿⣿⢸⡿⢃⠾⢿⠸⣿⣿⡿⣿⡧⠱⣜⠻⣿⣦⡙⢸⣿⣿⠀⠀⡀⠀ ⣠⣾⣿⡿⣿⣿⢈⣴⣿⣿⣿⣷⣝⢿⣿⡄⢭⣓⠉⠛⠊⠉⣛⢓⣈⣿⣀⠈⠃⠀ ⢛⣿⣿⡇⢿⣿⡜⠟⠛⠛⠿⣿⣿⣷⣯⣛⣃⢀⣴⣦⢠⡄⠈⡾⢣⣿⣿⣆⠀⠀ ⠿⠟⠋⢡⣌⠻⢿⠌⢁⣠⣤⣬⣿⣿⣿⣿⣿⡰⠿⠎⣼⡿⢀⣶⠟⠻⣿⣿⣦⡀ ⣌⢀⣔⢿⣿⠀⣄⠺⣿⡿⠿⣿⣿⡻⡟⠉⣿⣿⡿⠻⣿⠗⣠⠀⣾⡗⢈⣛⠛⢛ ⣱⡿⢿⣦⡉⠀⢈⣥⣴⣼⠃⠛⠛⠛⠶⠚⠛⠛⠀⢷⣶⣦⡅⠀⠛⣷⣿⣿⣷⣬ ⠉⣴⣿⡟⠁⠀⠈⠿⠿⠟⣃⣤⡀⠀⠀⠀⢠⣤⣬⠛⠿⠟⠃⠀⠀⠻⣿⣿⡉⣛ ⢸⣿⣿⠀⡆⠀⠀⠀⠀⠀⠉⢉⠁⠀⠛⠛⠀⠉⢉⡄⠀⠀⠀⠀⣰⠆⣿⣿⡷⢹
import sys, os from dataclasses import dataclass, field @dataclass class DeveloperProfile: name: str = "Ignaz Bhatta" academic_base: str = "BSc CS with AI · first year" location: str = "Kathmandu, Nepal" os_environment: str = "CachyOS (Arch) + Hyprland + Wayland" desktop_shell: str = "Quickshell / QML custom rice" active_models: list = field(default_factory=lambda: ["Qwen 3.8 27B", "Qwen 3.6 35B A3B"]) runtime: str = "llama.cpp / GGUF local quantization" print(DeveloperProfile())
# Private offline LLM inference pipeline $ ./llama-cli \ -m ./models/qwen3.8-27b-instruct-q5_k_m.gguf \ -p "System: You are running on Ignaz's local hardware rig in Kathmandu." \ --threads 16 \ --ctx-size 8192 \ --temp 0.2 [llama.cpp] Context loaded in 118ms. VRAM allocation optimized. Ready for local queries.
Hackathons, a rice that refuses to idle, and a lab with no API keys.
Decision engine for schools and NGOs to judge EdTech spend against infrastructure, readiness, and cost before anything is bought.
Kasthamandap A Levels hackathon project: a textured 2D map for placing landmarks and measuring disaster impact as a playable sim.
Daily-driver Hyprland on Arch/CachyOS. Widgets in Quickshell/QML, keybinds that stay in muscle memory, compositor chrome that disappears.
Offline inference on consumer silicon. Quantized GGUF, context cache offload, no cloud round-trip.
Click a tile. Steam pages where they exist, official sites everywhere else.
Kernel, compositors, low-level config
Software and automation
Open weights, local first
Where I study — not a sponsor reel
BSc (Hons) Computer Science with AI · First year (current) · Kathmandu
10+2 Science · Physics & Mathematics
SEE / Grade 10
Local AI setups, hackathon teams, Hyprland dots, open-source — inbox is open. Kathmandu, Nepal.
