Lars Grammel @lgrammel Twitter profile

Last Seen Profiles

@IcnixEsports

@Motumbo_96x

@gabfrancooo

@ZoeKinsella3

@GiantsXI

@Dias_6x

@CElbracko

@ConnieAnnKirk

@RezzyAU

@alitozaki

@EmoTyleer

@GabyNR

@Mvssi_x

@angell23xoxo

@TarantoC5eSport

@ayako91342385

@ArechiEsports

@v2Matty

@yarpolyanitsa

@alessiasmccabe

@st_Alfano9

@buzzisbackCUFC

@Buizilla

@FCSoulMedusa

@Unass_IdF

@IXTHUSR

@RedlineWusky

@Lauren9Dudley

@Martha372013644

@UgtJosep

@Opt_In_Props

@_DSAG

@Tousadroite

@fripujols

@stw_pdg

@Persephiroth

Lars Grammel

@lgrammel

6 months

Function calling with ModelFusion & OpenHermes 2.5 @Teknium1 🚀 This is the first time automatic function selection & calls work well for me with open source 7B models. Super excited about the potential for assistants that are powered by open source models. @Ollama_ai example:

4

19

231

Lars Grammel

@lgrammel

2 months

Today is my first day at @vercel - super excited to help build the AI SDK!

29

2

187

Lars Grammel

@lgrammel

1 year

Rubberduck is an AI chatbot for developers. Here are a few ways how Rubberduck takes ChatGPT-like chat interfaces to the next level 👇

3

13

154

Lars Grammel

@lgrammel

1 year

Introducing Rubberduck, the chat assistant that makes coding easier! Get help with code explanations, error diagnosis, and more, all within your IDE. Sign up for the alpha waitlist to try it out:

3

12

137

Lars Grammel

@lgrammel

1 month

We shipped four new experimental functions in the AI SDK v3.0.15 that simplify calling LLMs: ◆ Text generation and streaming, including tool calling ◆ Structured object (JSON) generation and streaming You can use them with our OpenAI and Mistral providers.

11

12

152

Lars Grammel

@lgrammel

1 year

GPTAgent.js v0.0.6 🤖 Example implementation of @yoheinakajima 's BabyAGI in TypeScript New model: gpt-3.5-turbo support #BabyAGI #typescript

7

5

85

Lars Grammel

@lgrammel

1 month

Vercel AI SDK v3.0.19 adds tool use to the ai/core Anthropic provider. You can use it with generateText and generateObject:

Anthropic

@AnthropicAI

1 month

Tool use is now available in beta to all customers in the Anthropic Messages API, enabling Claude to interact with external tools using structured outputs.

70

288

2K

7

8

86

Lars Grammel

@lgrammel

30 days

Vercel AI SDK v3.0.20 moves the following features out of experimental: ◆ StreamData ◆ useAssistant & AssistantResponse It also adds streaming support to AWS Bedrock/Cohere. Here's an updated example useCompletion route for the Next.js app router:

5

4

84

Lars Grammel

@lgrammel

15 days

You can use Vercel AI SDK Core functions with OpenAI-compatible providers such as Groq, Perplexity and Fireworks. Here are a few example chat routes for the Next.js app router with streamText:

7

6

73

Lars Grammel

@lgrammel

1 month

Vercel AI SDK v3.0.18 adds an Anthropic AI provider that can be used with streamText and generateText (only messages API, no tool calls yet). Here's a multi-modal text streaming example:

6

5

65

Lars Grammel

@lgrammel

6 months

Wrapped up my first week contributing to the @vercel AI SDK. Super exciting project - great UI integrations and client/server interactions will take AI apps to the next level.

2

61

Lars Grammel

@lgrammel

6 months

The @vercel AI SDK supports ModelFusion streaming out of the box 🤯 Here is an example of using @Ollama_ai / Llama 2 chat with the AI SDK and ModelFusion:

1

3

53

Lars Grammel

@lgrammel

5 months

I've added support for the @ollama chat API to ModelFusion. The API makes creating chatbots much easier! Here's a complete console chatbot example:

2

8

45

Lars Grammel

@lgrammel

27 days

Update for AI SDK v3.0.22: The providers for the experimental AI Core functions (OpenAI, Mistral, Google, Anthropic) are now split into separate packages within the ai-sdk NPM org. You need to install the necessary packages to continue using these functions. Example usage:

3

1

44

Lars Grammel

@lgrammel

17 days

The latest AI provider versions for the Vercel AI SDK simplify calling LLMs further. You don't need to specify the API via .chat() or .completion() any more when you use them with generateText, streamText, generateObject or streamObject:

6

1

43

Lars Grammel

@lgrammel

4 months

@abacaj I wonder if we truly want agents with full autonomy or if we actually want more restricted automation (that can still involve LLMs, RAG and so forth), especially in a business context?

3

0

41

Lars Grammel

@lgrammel

28 days

With OpenAI GPT-4-Turbo you can now use tools and multi-modal inputs at the same time. You can leverage this feature to derive structured data directly from images. Here is an example that uses generateObject from the AI SDK to describe and critique AI-generated art:

0

2

40

Lars Grammel

@lgrammel

12 years

#stackoverflow documentation for #android and #java matches the actual API usage: http://t.co/Bdfp1Owy http://t.co/gcwuSjDt

3

45

33

Lars Grammel

@lgrammel

1 month

AI SDK v3.0.17 adds a Google Generative AI provider that can be used with streamText and generateText (incl. tool calls). Here's a command line gemini-pro chatbot:

3

32

Lars Grammel

@lgrammel

14 days

The OpenAI provider for the Vercel AI SDK v0.0.7 supports object generation and streaming for @FireworksAI_HQ Here are examples that use firefunction-v1:

3

35

Lars Grammel

@lgrammel

5 months

🎉 My article on creating a local chatbot using @ollama , Next.js, and ModelFusion has been featured on @hashnode ! Check it out here:

Create Your Own Local Chatbot with Next.js, Ollama, and ModelFusion

In this blog post, we'll build a Next.js chatbot that runs on your computer. We'll use Ollama to serve the OpenHermes 2.5 Mistral LLM (large language model) locally, the Vercel AI SDK to handle...

p42.hashnode.dev

1

5

29

Lars Grammel

@lgrammel

1 year

In the next few days, I'll be building an open source chatbot extension for #vscode that uses #OpenAI #GPT . It'll be able to explain code, diagnose errors, and more. You can follow along in this twitter thread. #buildinpublic

1

4

28

Lars Grammel

@lgrammel

6 months

Guide for running the BakLLaVA Vision LLM ( @far__el ) with llama.cpp ( @ggerganov ) and ModelFusion🤖👀

Lars Grammel

@lgrammel

6 months

🚀 Exciting update! ModelFusion now supports multi-modal inputs (text & images) through llama.cpp @ggerganov ModelFusion + BakLLaVA in action 🎥👇

1

0

6

0

2

23

Lars Grammel

@lgrammel

10 months

@SullyOmarr @vercel I've been hacking on something similar in my spare time, bc I wanted something vendor-neutral. Would love to hear what you think.

GitHub - vercel/modelfusion: The TypeScript library for building AI applications.

The TypeScript library for building AI applications. - vercel/modelfusion

github.com

1

2

22

Lars Grammel

@lgrammel

4 months

🧪With LLMs, you can easily generate fully typed, structured data, e.g. for tests, webpage content, and database seeding. My latest blog explores using @ollama and @Teknium1 's Nous-Hermes 2 Mixtral alongside Zod and ModelFusion for this use case. Here is the example code:

1

4

20

Lars Grammel

@lgrammel

10 months

@minimaxir @abacaj If you’re using JS/TS and want to have a nicer API for LLM calls (and a bit more), check out

GitHub - vercel/modelfusion: The TypeScript library for building AI applications.

The TypeScript library for building AI applications. - vercel/modelfusion

github.com

1

3

20

Lars Grammel

@lgrammel

3 years

@curtiseinsmann Beautiful comment, explains the regexp so well 🤩 When you come across a complicated regexp without such a comment, using a tool like can help:

1

19

Lars Grammel

@lgrammel

4 months

My take is that eventually almost all apps will include AI features. AI engineering will just be part of building apps, and it'll often happen in the language that the app is being built in. And many web apps are built in JS/TS.

Hrishi

@hrishioa

4 months

Controversial bet: JS/TS will win over python in the LLM space. Not for orchestrating models, but for application building. Quick reasons: 1. Easier packaging and delivery 2. LLM infra+orchestration will stabilize. Once you can run models efficiently with guided generation, you…

24

10

121

2

5

19

Lars Grammel

@lgrammel

3 months

@nisten @abacaj Not sure about that. I compared mistral-medium through the Mistral API with miqu-q5 using llama.cpp. They are pretty close for what I test (thought not exactly the same, and I didn't test code).

1

19

Lars Grammel

@lgrammel

1 month

First use of streamObject in the wild 🤩

GitHub - miurla/morphic: An AI-powered search engine with a generative UI

An AI-powered search engine with a generative UI. Contribute to miurla/morphic development by creating an account on GitHub.

github.com

Yoshiki Miura

@miiura

1 month

Introducing morph: a fully open-source AI-powered answer engine with a generative UI. Built with @vercel AI SDK, it delivers awesome streaming results. 👇 More details

90

193

1K

3

0

18

Lars Grammel

@lgrammel

6 months

I've added the ChatML prompt format to ModelFusion to support Open Hermes 2.5 by @Teknium1 Open Hermes 2.5 is currently on 2nd place in the @huggingface Open LLM leaderboard for the 7B class - it's a really good model that you can easily run on an M1 MacBook.

Teknium (e/λ)

@Teknium1

6 months

Today I am releasing Open Hermes 2.5! This model used the Hermes 2 dataset, with an added ~100k examples of Code Instructions, created by @GlaiveAI ! This model was originally meant to be OpenHermes-2-Coder, but I discovered during the process that it also improved almost every…

48

137

1K

0

1

17

Lars Grammel

@lgrammel

6 years

And last but not least, we've moved into our new @trifacta office in Berlin, and are looking for software engineers, customer solution architects, and customer success managers:

0

6

16

Lars Grammel

@lgrammel

6 months

Documentation for using the Vercel AI SDK, ModelFusion & Ollama:

Lars Grammel

@lgrammel

6 months

The @vercel AI SDK supports ModelFusion streaming out of the box 🤯 Here is an example of using @Ollama_ai / Llama 2 chat with the AI SDK and ModelFusion:

1

3

53

1

3

16

Lars Grammel

@lgrammel

8 years

Trifacta is looking for a #UX designer in Berlin - design the future of working with data!

0

9

14

Lars Grammel

@lgrammel

6 months

@morgymcg @Teknium1 @jxnlco it’s much weaker than gpt-4 in my experience

Lars Grammel

@lgrammel

6 months

In my experiments w/ parallel function calls, I used gpt-4-1106-preview to solve middle-school math problems. Previously GPT-4 got all of them right. The new gpt-4-1106-preview does not. And I'm not the only one noticing this quality drop:

0

11

0

1

14

Lars Grammel

@lgrammel

4 months

Here's how I think about the software stack for LLM inference, from a JS/TS dev point of view: There are 6 levels that build on one another: 1) The model: the actual model that will be executed at inference time. Sometimes it's the providers models (e.g. GPT-4 et al for…

2

0

14

Lars Grammel

@lgrammel

3 months

TDD: Twitter Driven Development When you just add the next hacky thing to your app to get likes & retweets.

1

0

13

Lars Grammel

@lgrammel

3 years

@catalinmpit What are you looking for in particular? I've 30+ years of programming experience and I've worked / am working with all those technologies, but I usually just figure stuff & get it done as I need to - so don't necessarily have the depth that might be required for this.

0

13

Lars Grammel

@lgrammel

4 months

Some thoughts on LLM APIs and client libraries: Right now there is a trend that every LLM provider (OpenAI, Anthropic, Google, Mistral, Ollama...) has a unique API and is implementing client libraries for Python, JS, etc. This makes a lot of sense for branding reasons and to…

2

13

Lars Grammel

@lgrammel

7 months

🎉 ModelFusion has just hit 500 stars on GitHub! 🎉 Thanks to everyone for the incredible support. Your contributions and feedback are driving this project forward! Explore the current state of ModelFusion: #OpenSource #AI #TypeScript #MachineLearning

1

2

13

Lars Grammel

@lgrammel

6 months

Next.js, @vercel AI SDK, Llama.cpp (by @ggerganov ), and ModelFusion starter with examples for: - Llama 2 chat - Mistral 7B ( @MistralAI ) - OpenHermes 2.5 ( @Teknium1 )

2

1

12

Lars Grammel

@lgrammel

3 years

@afilina @KentBeck Oh yes, so true! It also means that the situation will only get worse, not better, since the developers won't have the time to refactor and clean up technical debt.

2

0

12

Lars Grammel

@lgrammel

1 year

✨Rubberduck for VS Code 1.0✨ Use AI-powered code edits, explanations, test generation, error diagnosis, and chat with the official OpenAI API. Rubberduck is open source! 👇

1

4

12

Lars Grammel

@lgrammel

6 months

In my experiments w/ parallel function calls, I used gpt-4-1106-preview to solve middle-school math problems. Previously GPT-4 got all of them right. The new gpt-4-1106-preview does not. And I'm not the only one noticing this quality drop:

Chat GPT4 1106 vs ChatGPT 4: Impressive drop in quality

Hello, I’ve been doing a lot of testing since yesterday and so far the new version ChatGPT 4 0611 is really not as good in terms of response quality as Chat GPT4. Any ideas for improving the quality...

community.openai.com

0

11

Lars Grammel

@lgrammel

1 month

You can find out more in our docs: The new functions are currently experimental - we'd love to get your feedback!

Vercel AI SDK

Build AI-powered applications with the latest AI language models

sdk.vercel.ai

1

11

Lars Grammel

@lgrammel

8 months

ModelFusion v0.26.0 released 🎉 Added @lmnt_com text-to-speech integration. With LMNT, you can transform your text into dynamic and captivating speech in the blink of an eye! 🚀🔊 Learn more:

0

5

11

Lars Grammel

@lgrammel

8 years

The weird things humans do with spreadsheets #csvconf

0

4

11

Lars Grammel

@lgrammel

1 year

@JayScambler @yoheinakajima I wrote yet another TypeScript version of BabyAGI:

Lars Grammel

@lgrammel

1 year

GPTAgent.js v0.0.6 🤖 Example implementation of @yoheinakajima 's BabyAGI in TypeScript New model: gpt-3.5-turbo support #BabyAGI #typescript

7

5

85

2

1

11

Lars Grammel

@lgrammel

6 months

Function call prompt inspired by @abacaj with a few improvements that increased robustness for me:

anton

@abacaj

6 months

Some folks were asking how to do the function calling with openhermes-2.5, notebook is posted here now

15

28

274

1

11

Lars Grammel

@lgrammel

6 months

🚀 New in ModelFusion: @Ollama_ai support! Use Ollama's local text generation, streaming, and embedding within your ModelFusion apps. Text streaming:

1

2

11

Lars Grammel

@lgrammel

6 years

Google Cloud Dataprep by @trifacta , the product I've been working on for the past 2 years, has been updated with many new features and a design refresh:

Home

Alteryx powers actionable insights; more than 8,000 customers rely on Alteryx to automate analytics to improve revenue performance, manage costs, and mitigate risks.

www.alteryx.com

1

5

11

Lars Grammel

@lgrammel

4 years

@jankeromnes @tlakomy Yes! I totally agree on the cost of unit tests, with one exception: they are great for testing algorithms with clearly defined, small interfaces. I usually go with the guideline from

0

10

Lars Grammel

@lgrammel

4 months

@rohanpaul_ai For JS/TS, you can use llama.cpp and ModelFusion generateStructure. It automatically generates and uses GBNF grammars to constrain the output.

0

10

Lars Grammel

@lgrammel

8 years

I'm looking for #devops engineers to join the @trifacta team in #Berlin - work with #gcp , #aws , #hadoop and more

0

16

9

Lars Grammel

@lgrammel

4 months

Streaming AI-generated, structured data w/ React & Ollama:

1

0

9

Lars Grammel

@lgrammel

1 month

@SullyOmarr @miiura @vercel Here’s an example:

2

1

10

Lars Grammel

@lgrammel

3 years

@oliverjumpertz @svpino Very true! However, if you don't ask for growth opportunities, or to work on areas/projects you want to work on, or to do open source or whatever it is - your manager / the company will not read your mind and give it to you. It's very important to be proactive and ask imo.

1

0

10

Lars Grammel

@lgrammel

3 months

I added an object generator tool to ModelFusion that creates structured outputs using JSON schemas. It could be quite useful for generating fictional data, e.g. for RPGs. Here is an example that generates enemies in a fantasy world:

2

1

10

Lars Grammel

@lgrammel

4 years

Love this idea! Here are the software engineering books from my shelf that I’d recommend:

Mario Cervera

@macerub

4 years

📚 This is my shelf of technical books. I would recommend all of them. They contain invaluable knowledge. ❓ Do you have any question about any of these books? If you do, you can leave me a comment in this tweet. I promise that I will do my best to answer 🙂

27

33

255

4

0

10

Lars Grammel

@lgrammel

4 months

Streaming AI-generated objects with Next.js 14 & OpenAI gpt-4-turbo:

2

1

10

Lars Grammel

@lgrammel

7 years

Very excited and proud that Cloud Dataprep is now publicly available:

Trifacta is now part of Alteryx—analytics for all!

@Trifacta

7 years

Built in collaboration w/ Trifacta, @googlecloud Dataprep is now available in public beta! Explore the features:

0

23

30

0

1

9

Lars Grammel

@lgrammel

5 months

✨Introducing WhisperWriter ✨ A local & private voice controlled transcription panel Using Next.js, whisper.cpp, ModelFusion, and voice activity detection. (turn on audio; the transcriptions are better than what's in the video if you use stronger models)

1

0

8

Lars Grammel

@lgrammel

4 months

Flow engineering is here to stay. It is about communicating context and applying constraints to LLMs to achieve the desired outcome for a use case, and is an integral part of AI engineering.

Andrej Karpathy

@karpathy

4 months

Prompt engineering (or rather "Flow engineering") intensifies for code generation. Great reading and a reminder of how much alpha there is (pass @5 19% to 44%) in moving from a naive prompt:answer paradigm to a "flow" paradigm, where the answer is constructed iteratively.

125

545

3K

0

1

9

Lars Grammel

@lgrammel

4 months

@ggerganov I've revamped the JS/TS client for llama.cpp server in ModelFusion. It is afaik the most complete client that's available for the llama.cpp server in JS/TS. It includes: - text generation - text streaming - structure generation with automatic gbnf creation from json…

2

0

8

Lars Grammel

@lgrammel

3 years

Learning tip New libraries, frameworks and programming languages come out *all the time*. A solid understanding of software engineering concepts is the key to learning them quickly - focus on the foundations that rarely change and the rest will be much easier!

0

3

8

Lars Grammel

@lgrammel

14 years

Great intro to #CSS resets: http://bit.ly/9zhzKP

0

3

8

Lars Grammel

@lgrammel

5 months

Added @MistralAI support to ModelFusion 🚀

1

0

8

Lars Grammel

@lgrammel

11 months

New JS library I've been working on for the past few weeks (it's still early / experimental): It takes care of some of the annoying stuff when calling model APIs (e.g., retries, logging, throttling) and provides a simpler API:

1

0

8

Lars Grammel

@lgrammel

3 years

Tech debt is such a negative metaphor. We could also think about removing tech debt as tech investment, and selling the benefits it creates (such as increased velocity, code quality, developer morale & retention, ability to hire) - without even calling it debt 🤔

3

1

8

Lars Grammel

@lgrammel

4 months

The @perplexity_ai online LLMs are brilliant. Up-to-date information from the web, exposed through a standard LLM interface. And the speed of their RAG is impressive. Gamechanger, especially for news.

2

0

7

Lars Grammel

@lgrammel

6 years

I'm very excited that August is starting off with lots of announcements & goodness at @trifacta

1

0

8

Lars Grammel

@lgrammel

3 years

@catalinmpit GKE Kubernetes clusters are also down it seems. Fun Monday.

1

0

8

Lars Grammel

@lgrammel

8 years

Want to work on a #javascript app that redefines working w/ data? Apply to @trifacta #Berlin

0

9

8

Lars Grammel

@lgrammel

9 months

I always enjoy the talks at the Berlin AI meetups. Today @langfuse was presenting on metrics and observability for LLMs.

0

1

7

Lars Grammel

@lgrammel

1 year

⚡introducing chatbot.js ⚡ Setup your own AI chatbot HTTP server in less than 3 minutes. You can configure a chat template and access past messages. It's super basic atm, but it'll grow 🌱 #javascript #openai #NodeJS

GitHub - lgrammel/chatbot.js: A simple chatbot service

A simple chatbot service. Contribute to lgrammel/chatbot.js development by creating an account on GitHub.

github.com

0

1

7

Lars Grammel

@lgrammel

3 months

@nisten @abacaj For the mars calculation task I'm getting different results in llama.cpp (max 1024 tokens). Not saying that this is mistral-medium, but the difference seems to be closer than what I thought after seeing your tweet.

0

1

7

Lars Grammel

@lgrammel

3 months

Wrote a tutorial on how to stream GPT-4 generated, structured content in Next.js:

Next.js and GPT-4: A Guide to Streaming Generated Content as UI Components | ModelFusion

Next.js and GPT-4 - A Guide to Streaming Generated Content as UI Components

modelfusion.dev

1

0

6

Lars Grammel

@lgrammel

9 months

Introducing ModelFusion: Build AI apps with JavaScript and TypeScript. ModelFusion is a toolbox, not a limiting framework. It's designed to grant you the full power and control over calling the AI models while keeping things clear and accessible.

Introducing ModelFusion | ModelFusion

Open-source library for building AI applications, chatbots, and agents with JavaScript and TypeScript.

modelfusion.dev

0

1

6

Lars Grammel

@lgrammel

5 months

New blog post: Create your own local chatbot with Next.js, Ollama, OpenHermes 2.5 Mistral, the Vercel AI SDK, and ModelFusion.

Create Your Own Local Chatbot with Next.js, Ollama, and ModelFusion

In this blog post, we'll build a Next.js chatbot that runs on your computer. We'll use Ollama to...

dev.to

0

6

Lars Grammel

@lgrammel

3 years

@oliverjumpertz Great tip! Got inpired and added it to the VS Code extension I'm developing - now I get automatically reminded about the exponentiation operator :)

1

0

7

Lars Grammel

@lgrammel

3 years

Love the current tech stack that I'm using: - Language: @typescript - Frontend @reactjs @MaterialUI #nextjs - Server: #nextjs #Kotlin - Backend: #graphql @HasuraHQ @PostgreSQL - Deployment on @kubernetesio @GCPcloud Most of it has been invented in the last 5-10 years 🤯

0

6

Lars Grammel

@lgrammel

3 years

I'm pretty happy with how powerful P42 is becoming. Here's an example of refactoring a small JS function using only automated code actions (no error-prone manual changes). I chain together 11 P42 refactorings and one VS Code cleanup to arrive at a much simpler function:

2

6

Lars Grammel

@lgrammel

3 years

P42+ adds repository analysis, automated pull request and commit review, and bulk changes for folders to complement the @p42ai JavaScript Assistant functionality. If you are using #GitHub , give it a try! It's free for public open-source projects. #javascript #typescript

P42 JS Assistant

@p42ai

3 years

Since the first private beta release exactly a year ago, P42 has grown from a code modernization tool for #GitHub to a programming assistant for #JavaScript and #TypeScript . P42 for GitHub has received a significant updated and is now P42+. Learn more:

0

1

4

0

3

5

Lars Grammel

@lgrammel

11 years

~"Given the starvation potential Node isn't a solution for CPU intensive problems. It's a solution for I/O problems." http://t.co/uvlnd4Qqvf

0

2

6

Lars Grammel

@lgrammel

6 months

🚀 ModelFusion now integrates SQLite as a vector store through SQLite VSS by @agarcia_me A huge shoutout to @experilearning for their amazing contribution to this extension. 🙌 Learn more:

0

6

Lars Grammel

@lgrammel

2 years

Are you working on #JavaScript or #TypeScript projects in #vscode ? Here are 6 extensions that improve your editing experience with AI-generated code completions, automated formatting, linting, spell checking, and automated refactorings:

1

3

4

Lars Grammel

@lgrammel

4 years

@shreyas @giridharvc7 My experience: WAYRTTD is great to focus on a core issue - the 5 Whys can help identify layers of a problem and invest in fixing the problem at all those layers (e.g. in software engineering, fix the bug, improve logging to find it earlier, and improve the testing).

0

1

6

Lars Grammel

@lgrammel

3 years

Tired of modernizing your JavaScript and TypeScript code manually? I've created a codemod tool that helps you use new language features and apply best practices. You can directly open GitHub pull requests with suggestions. Check out P42: #buildInPublic

1

3

6

Lars Grammel

@lgrammel

10 years

great analogy by @jeffrey_heer "encounters with new data should bring the child-like excitement of opening a present" http://t.co/8NdEjWn4FR

1

2

6

Lars Grammel

@lgrammel

1 year

JS Agent v0.0.15 - New PDF summarizer example with GPT-4 and extract & rewrite pipeline - PDF loading - Better text extraction prompts - Changed summarize to extract text #ChatGPT #buildinpublic

GitHub - lgrammel/js-agent: Build AI Agents with JS & TS

Build AI Agents with JS & TS. Contribute to lgrammel/js-agent development by creating an account on GitHub.

github.com

0

6

Lars Grammel

@lgrammel

8 years

Care about data tools and quality? We're looking for a quality engineer @trifacta #Berlin

0

6

4

Lars Grammel

@lgrammel

6 months

Function calling with OpenHermes 2.5, llama.cpp @ggerganov , and ModelFusion:

1

0

6

Lars Grammel

@lgrammel

6 months

🚀 Exciting update! ModelFusion now supports multi-modal inputs (text & images) through llama.cpp @ggerganov ModelFusion + BakLLaVA in action 🎥👇

1

0

6

Lars Grammel

@lgrammel

1 year

Early module dependency visualization prototype that I've started to work on:

1

6

Lars Grammel

@lgrammel

9 months

@izzyz This complexity annoyed me so much that I ended up writing an alternative for TS/JS:

GitHub - vercel/modelfusion: The TypeScript library for building AI applications.

The TypeScript library for building AI applications. - vercel/modelfusion

github.com

1

0

6

Lars Grammel

@lgrammel

4 years

Great insights on the shift to remote work this decade. Corona triggered it, but the new way of working (and living) is here to stay.

Chris Herd

@chris_herd

4 years

I've spoken to around 1,000 companies over the last 6 months about their plans for remote work going forward Here are a few things I've learned [ a thread ] 💻🏠🌍

435

5K

11K

2

6

Lars Grammel

@lgrammel

1 year

Chatting with an AI right next to your code is a pretty exciting experience. Building this feature was fun, and to my surprise, I started enjoying trying to understand unfamiliar code more, even when the AI answers were not always accurate.

P42 JS Assistant

@p42ai

1 year

✨ JS Assistant 2.10✨ New: Chat with AI about code explanations

0

1

3

0

6

Lars Grammel

@lgrammel

3 years

@MattCodeJourney I very much recommend vega & vega-lite

Vega: A Visualization Grammar

Vega is a declarative format for creating, saving, and sharing visualization designs. With Vega, visualizations are described in JSON, and generate interactive views using either HTML5 Canvas or SVG.

vega.github.io

1

0

6

Lars Grammel

@lgrammel

4 months

@abacaj Yeah. I've implemented that in some limited fashion in business contexts, and it works fairly well. Flow engineering will eventually evolve into that direction for some harder problems imo.

1

0

6

Lars Grammel

@lgrammel

6 years

We've upgraded our Trifacta Wrangler product with the latest and greatest features:

1

0

6