lmarsden Profile Banner
Luke Marsden Profile
Luke Marsden

@lmarsden

Followers
3K
Following
2K
Media
526
Statuses
4K

CEO @helixml - Private GenAI Platform https://t.co/5SOOUrIadv. Owner https://t.co/IfLLD1AyIH. Founder https://t.co/HT0TgMQTIh. Hacker & entrepreneur

London, England
Joined February 2009
Don't wanna be here? Send us removal request.
@arkhai_io
Arkhai
22 days
Ten thousand years of human commerce just became the beta version. v1.0 ships today. Introducing Arkhai, where intelligence meets agency.
29
45
142
@lmarsden
Luke Marsden
3 months
@zeddotdev @zeddotdev you need to fix your summarization at long context lengths though
0
0
0
@lmarsden
Luke Marsden
4 months
@zeddotdev is multiple concurrent threads something you're working on?
1
0
0
@lmarsden
Luke Marsden
4 months
Just tried @zeddotdev out of frustration at how the Cursor UI would grind to a halt (especially with inline terminals with lots of output) and holy cow, those guys are nailing it. I'm just not sure multiple concurrent threads are working...
1
0
4
@lmarsden
Luke Marsden
4 months
*We
0
0
1
@lmarsden
Luke Marsden
4 months
Deploy here!
0
0
1
@lmarsden
Luke Marsden
4 months
Check this out, then deploy Helix on 2x A100 80GB GPUs with $5, a few clicks, and a 30 minute wait. Then deploy and build agents on it! BTW, we just hit $1M annual revenue
@CloudcastPod
The Cloudcast - Cloud + AI
4 months
Building Private GenAI stacks https://t.co/tLMjZxvnQg Luke Marsden (@lmarsden, CEO @HelixML) talks about Private GenAI. What is it? Why do you need it? We also discuss the layers of a Private GenAI Stack, and why most organizations are opting for RAG over fine-tuning LLMs.
1
0
1
@lmarsden
Luke Marsden
4 months
I finally made a profitable software company
3
0
14
@DrPhilWinder
Phil Winder
5 months
Kodit 0.3 Released: - 10× faster indexing - Private Azure DevOps support - Pre-filter searches - Auto-indexing - Slick CLI progress bars https://t.co/JxvY3dfr6e
Tweet card summary image
blog.helix.ml
Information about Kodit's latest 0.3 release
0
1
2
@lmarsden
Luke Marsden
6 months
Here we make our first foray into coding assistants - with an **open source** MCP server that indexes many private repos, making your code assistant smart when it comes to private, internal enterprise coding conventions. Can't wait to integrate this into @helixml
@DrPhilWinder
Phil Winder
6 months
I've found that AI coding assistants struggle with new projects, private codebases, library updates, etc. So I built Kodit, an MCP server to index external repositories: https://t.co/R4S2mFevdB I'm looking for early feedback. Check it out and let me know! Thanks!
0
1
2
@lmarsden
Luke Marsden
6 months
Here we make our first foray into coding assistants - with an **open source** MCP server that indexes many private repos, making your code assistant smart when it comes to private, internal enterprise coding conventions. Can't wait to integrate this into @helixml
@DrPhilWinder
Phil Winder
6 months
I've found that AI coding assistants struggle with new projects, private codebases, library updates, etc. So I built Kodit, an MCP server to index external repositories: https://t.co/R4S2mFevdB I'm looking for early feedback. Check it out and let me know! Thanks!
0
1
2
@lmarsden
Luke Marsden
7 months
Can't wait to run this in @helixml so you can GitOps without sending your shit to @sama
@stefanprodan
Stefan Prodan
7 months
We've released the MCP Server for @fluxcd 🚀🤖 🚀 https://t.co/nYH34xC9Rg
0
0
1
@lmarsden
Luke Marsden
7 months
0
0
0
@lmarsden
Luke Marsden
7 months
Accelerating inference for latent diffusion models?
1
0
0
@Ben_Hall
Ben Hall
8 months
Team @helixml at #KubeCon
1
1
1
@lmarsden
Luke Marsden
9 months
LOOK AT THE DATA
@HamelHusain
Hamel Husain
9 months
. @sh_reya 's paper confirms what I see in practice 1) Automated evals don't work (without semi-manual human alignment) 2) Most tools don't provide this alignment 3) Automated evals add mostly noise 4) You can only write good evals by looking at data and reacting to failures
0
0
0
@HamelHusain
Hamel Husain
9 months
. @sh_reya 's paper confirms what I see in practice 1) Automated evals don't work (without semi-manual human alignment) 2) Most tools don't provide this alignment 3) Automated evals add mostly noise 4) You can only write good evals by looking at data and reacting to failures
19
80
711
@lmarsden
Luke Marsden
9 months
Please use your $480M to buy some more GPUs so I don't have to keep testing on 8xA100s in Sydney!
0
0
1