How you find out whether an agent is any good: held-out splits, trajectory-level scoring, and leaderboards that publish their method.
👋 Open to Work
Yauhen Bichel
YauhenBichel
·
AI & ML interests
Create new models, create harnesses, use projects with AI agentic apps, platform engineering
Recent Activity
updated a dataset 2 days ago
YauhenBichel/py-harness-intents updated a dataset 3 days ago
YauhenBichel/yserver-durability-experiments published a dataset 3 days ago
YauhenBichel/yserver-durability-experimentsOrganizations
Dermoscopic lesion classification — public data and models behind the CNN work I did at MoleCare (NHS App Library). Research, not clinical tools.
Models small enough to run on infrastructure you control — the practical constraint when data cannot leave the estate.
-
Qwen/Qwen3-4B-Instruct-2507
Text Generation • 4B • Updated • 3.64M • • 993 -
Qwen/Qwen3-4B-Thinking-2507
Text Generation • 4B • Updated • 568k • • 619 -
meta-llama/Llama-3.2-3B-Instruct
Text Generation • 3B • Updated • 1.62M • • 2.74k -
bartowski/Llama-3.2-3B-Instruct-GGUF
Text Generation • 3B • Updated • 151k • 242
How you find out whether an agent is any good: held-out splits, trajectory-level scoring, and leaderboards that publish their method.
Dermoscopic lesion classification — public data and models behind the CNN work I did at MoleCare (NHS App Library). Research, not clinical tools.
Agents that diagnose production incidents — the corpora, tooling and tool-calling models. Built around my work on OpenSRE's evaluation environment.
Models small enough to run on infrastructure you control — the practical constraint when data cannot leave the estate.
-
Qwen/Qwen3-4B-Instruct-2507
Text Generation • 4B • Updated • 3.64M • • 993 -
Qwen/Qwen3-4B-Thinking-2507
Text Generation • 4B • Updated • 568k • • 619 -
meta-llama/Llama-3.2-3B-Instruct
Text Generation • 3B • Updated • 1.62M • • 2.74k -
bartowski/Llama-3.2-3B-Instruct-GGUF
Text Generation • 3B • Updated • 151k • 242