menu

pawe jankes

@Pawe Jankes

Pawel Jankiewicz

Creator of SSI V5 | Universal Lab | Persistent AI agents and LLM optimization
github.com/jankes72/SSI_V5 Joined October 2026
121 Points • 2 Badges • 0 Connections • 0 Followers • 3 Following
Profile Stats Wall Posts Badges

About

I develop SSI V5, an independent research programme for persistent AI agents, adaptive routing and evidence-based evaluation. Universal Lab brings research conversations and bounde... Show more

Language & Tools

Python, Linux, Docker, Ollama, llama.cpp, GGUF, LLM evaluation and multi-agent systems

Currently Exploring

Connecting Research Memory, S10/V10 selection and SCC execution with bounded task generation. Quantization, calibration, LoRA and model memory/time trade-offs. Preparing an SSI configuration for 8 GB RAM systems.

Achievements

Published SSI MultiQuant CPU measurements: 3.14% smaller GGUF than a standard Q4_K_M control and about 1.82% lower sampled process peak RSS in a short test, with effectively unchanged elapsed time. Separate 8 GiB experiments completed 9 simple checks across Bonsai 27B, Qwen 32B and Llama 70B using file paging. These are bounded research results; practical quality and SSI runtime integration remain under evaluation.

Top Tags

User Activities

JanFebMarAprMayJunJulAugSepOctNovDec
Mon
Tue
Wed
Thu
Fri
Sat
Sun
Less More

Contributions
Joined: 10 hours (since Oct 21)
Full Name: Pawel Jankiewicz
Headline: Creator of SSI V5 | Universal Lab | Persistent AI agents and LLM optimization
About: I develop SSI V5, an independent research programme for persistent AI agents, adaptive routing and evidence-based evaluation. Universal Lab brings research conversations and bounded experiments into one workflow. SSI MultiQuant explores model memory, latency and task quality. My public GitHub shares architecture, protocols, benchmark results and their limits. My next goal is to prepare and validate an SSI model configuration for computers with 8 GB RAM, accepting longer response times where needed.

Public evidence and documentation: https://github.com/jankes72/SSI_V5
Location:
Website: https://github.com/jankes72/SSI_V5
Languges & Tools: Python, Linux, Docker, Ollama, llama.cpp, GGUF, LLM evaluation and multi-agent systems
Currently Exploring: Connecting Research Memory, S10/V10 selection and SCC execution with bounded task generation. Quantization, calibration, LoRA and model memory/time trade-offs. Preparing an SSI configuration for 8 GB RAM systems.
Achievements: Published SSI MultiQuant CPU measurements: 3.14% smaller GGUF than a standard Q4_K_M control and about 1.82% lower sampled process peak RSS in a short test, with effectively unchanged elapsed time. Separate 8 GiB experiments completed 9 simple checks across Bonsai 27B, Qwen 32B and Llama 70B using file paging. These are bounded research results; practical quality and SSI runtime integration remain under evaluation.
Fun Fact:
Random Dev Quote:
Latest Video:
chevron_left

Active in these Groups:

Angular Dev
289 members
CoderLegion Changelog
241 members
Laravel Developer
291 members