╭──────────────────────────────╮
│ 22elix3r │
│ build · break · understand │
╰──────────────────────────────╯
CS student from India. I like working close enough to the machine that things eventually become interesting.
Most of my time disappears into local AI, inference, Linux, developer tooling, computer vision, and generative media.
I care less about collecting technologies and more about building things that answer questions like:
- can this model run faster on consumer hardware?
- can this annoying developer workflow be turned into a proper tool?
- can ML solve something useful outside another benchmark?
- why does this work, and what happens if I break it?
~/now
├── local-llm
│ ├── llama.cpp
│ ├── ik_llama.cpp
│ ├── quantization
│ └── inference benchmarking
│
├── building
│ ├── rust tooling
│ ├── python
│ └── open-source experiments
│
├── generative-ai
│ ├── image + video models
│ ├── LoRA training
│ └── ComfyUI
│
└── learning
├── algorithms
├── systems
└── whatever rabbit hole looked interesting this week
Languages
Python Rust C Java Bash
AI / ML
PyTorch llama.cpp ComfyUI LoRA GGUF
Dev
Linux Git Docker FastAPI
local AI ████████████████████
developer tooling ██████████████████░░
open source ██████████████████░░
systems ████████████████░░░░
computer vision ████████████████░░░░
creative technology █████████████████░░░
I tend to bounce between software engineering and visual work. Sometimes that means debugging an inference backend. Sometimes it means training a video model. Occasionally those two worlds collide and produce something useful.
Making large models run faster, locally, on hardware they were not particularly designed to fit on.
If I find something worth fixing upstream, I would rather send the patch than keep the workaround in a folder forever.
while curious:
build()
break_something()
understand_it()