RK @itisrmk

Joined November 2015

Tweets

97
Followers

22
Following

154
Likes

24

RK @itisrmk

3 months ago

Claude Code UI. ChatGPT subscription underneath. GPT-5.4 running through a local Codex bridge. The interface stays. The model becomes swappable. #ClaudeCode #ChatGPT #GPT5 #AIEngineering #BuildInPublic

1 0 1 85 0

View Details

RK @itisrmk

3 months ago

Claude Code Unlocked. Qwen 3 brain. MLX on macOS. Local. The moat is shifting from the model to the interface. Keep the workflow. Swap the brain. #ClaudeCode #Qwen3 #Ollama #MLX #LocalLLM

0 0 1 98 0

View Details

RK @itisrmk

3 months ago

yup

0 0 0 17 0

View Details

RK @itisrmk

3 months ago

Built this after the Claude Code leak started blowing up. What stands out is not just Claude. It is the normalization of multi-runtime AI systems: subscription direct API compatible proxies local models cloud-native inference That shift is bigger than the screenshot. #ClaudeCode #AIInfra #LLMEngineering #MCP #AgenticAI #OpenAI #Anthropic #Bedrock #VertexAI #BuildInPublic

2 0 2 125 0

View Details

RK @itisrmk

3 months ago

!!!

Chaofan Shou @Fried_rice

3 months ago

Claude code source code has been leaked via a map file in their npm registry! Code: …a8527898604c1bbb12468b1581d95e.r2.dev/src.zip

3K 8K 48K 35.7M 41K

0 0 0 19 0

View Details

RK @itisrmk

3 months ago

@steipete @melvynx

0 0 0 19 0

View Details

RK @itisrmk

3 months ago

If you've ever had to retrain a model from scratch because it forgot old data -- clearn is for you. Wrap once. Train forever. github.com/itisrmk/clearn

0 0 0 13 0

View Details

RK @itisrmk

3 months ago

Try it now: ``` pip install clearn-ai ``` GitHub: github.com/itisrmk/clearn PyPI: pypi.org/project/clearn… HuggingFace demo: huggingface.co/rahulmk/clearn… Star the repo if this is useful. PRs welcome.

1 0 0 27 0

View Details

RK @itisrmk

3 months ago

Your neural network forgets everything it learned the moment you train it on new data. It's called catastrophic forgetting, and it's why most production ML pipelines are a mess of full retrains. I built an open-source library to fix it in 4 lines of code.