Artyom Lisitsyn

My LLM usage

Local

I believe that open-weight LLMs are important to consider in an academic landscape to ensure open access and understanding of the tools in use. Though less powerful than frontier models, one can nonetheless achieve surprising usefulness from running models locally on powerful personal devices. My setup:

• Device : Framework Desktop with AMD Ryzen™ AI Max+ 395 - 128GB
  • At the date of purchase (2026.05), this device cost ~$3,500
• Inference engine : kyuz0's AMD Llama.cpp Toolbox
• Models : I download models from HuggingFace using quantizations provided by Unsloth
  • Qwen3.8-27B-GGUF Q8_0
  • Qwen3.8-Flash-Next Q4_K_XL
  • Gemma-4-31B Q8_0
• Harness/Tools :
  • Opencode harness
  • Kagi's search API to give my local models tools to search the web

Search

It is common for search engines to integrate LLMs into their search results, or for some to forego search engines entirely in favor of chat-interfaces with LLMs. To avoid paying for this service with my data or attention, I use Kagi's search engine and assistant. Here is their webpage that markets the service.

Proprietary

I recently began to use Anthropic's models, though not yet as part of any projects.