Welcome to the [Gray Blog], a dynamic exploration of the cutting-edge world of [Generative AI] and [AI development]. Dive deep into innovative [AI applications] with contributions from industry leaders like [Sebastian Wehkamp], [Francisco Pereira], and [Aravind Putrevu]. This is where AI meets creativity, from the intricate realm of [LLMs] to the exciting narratives of [AI-driven "Choose Your Own Adventure" games]. Our insight-rich content spans everything from AI-native apps and fine-tuning approaches to groundbreaking solutions that enhance [workplace productivity] and develop captivating tools like an [AI Photobooth]. Whether you’re intrigued by the efficiency of [small models] or fascinated by AI's transformative power in [photography] and [gaming], this blog offers a trove of valuable info to inspire and inform the AI-savvy reader. Join us as we navigate the thrilling landscape of innovation, guided by real-world projects and hands-on demos, straight from the vibrant hubs of [Amsterdam] and [Silicon Valley].

From Article:


From Video:

Ronny Roeller - Classifying Data with LLMs.mp4 - Exploring unlabeled data classification with efficiency, I demonstrated using Spotify feedback by splitting it into requests, issues, and positives. Our app converts feedback into actionable insights through dynamic tags and automated classification, supporting product strategy.
Gaurav Chandrashekar - Summarise.live.mp4 - Gorov's project uses AI to summarize YouTube videos, providing key insights and chapters. It's available as a Chrome plugin for content-heavy users. Aiming for high quality, Gorov is open to user feedback and collaboration for further development.
Kanishk Garg Talk - Whisper Clipboard.mp4 - I'm Kanish, in Amsterdam for work, and I've developed Whisper Clipboard after attending AI Tinkers events. It's a stack-based tool that enhances language models by mixing questions with context, making data entry easier.
Lucas Meijer - I'm not a firefighter.mp4 - Developed an app to convert interviews into reports, facing challenges with misgendering due to model limitations. Attempts to guess gender from transcripts led to errors, like suggesting HR candidates as firemen. Key lesson: separate prompts from example data to prevent errors.
Aravind Putrevu - Fine tuning LLMs for function calling.mp4 - JSON schema is effective for complex code specs but falters with intricate parameters. Retaining base model capabilities during fine-tuning prevents knowledge loss. Quality data is crucial, and function-calling models enhance workflows and enterprise search.
Tomás Arribas - AI Photobooth.mp4 - An AI photobooth prototype, designed for a wedding, uses stable diffusion 1.5 to randomly modify photos without props. It runs locally on a laptop with an RTX 3070, employing face and depth masking with control nets, enhanced by the Imagine AI Room library.
Max Engelen - ScribeWizard and StockBot powered by Groq.mp4 - As a computer engineer at Grok, we specialize in end-to-end tech stack management, from chip design to cloud API access. Developers can freely access our APIs via console.grok.com, and we offer tools like Scribe Wizard and Stockbot on GitHub.
Stefan Nae - Mobi the workday AI health coach.mp4 - Stefan transitions from physics to computer vision in industry and police work, and develops a fitness app. Currently, the focus is on enhancing workplace productivity with scientifically-backed scheduled breaks and Slack integration for improved health and efficiency.
Laurynas Keturakis - Post-Postman_ End-to-End API Testing with GenAI.mp4 - I'm Laudinas, a product engineer at Fireplane. Discover how an API client can automate request pre-filling, integrate request data, diagnose errors, and generate test cases using language models with a 100-line express-like API, Hano. It's open-source on GitHub and supports OpenTelemetry.
Lucas Meijer - speedrun blooper reel of a year of ai app development.mp4 - After 15 years with Unity, ongoing frustrations over inefficiencies and tech limits exist despite AI advances. Apps help by automating tasks like transcription and invoice payments, simplifying communication and daily activities, but challenges and gaps persist.
Amir Houieh - What the f**k is an AI-Native app.mp4 - I am Amir, founder of a Rotterdam-based startup, simplifying AI for developers since 2018. We've developed a plug-and-play system for seamless AI integration, focusing on personalization and context, unifying frameworks for efficient app creation without extensive tech expertise.
Tijs Zwinkels - unicornsh A LLM in your (zsh) shell.mp4 - The LM terminal seamlessly integrates with the shell, enabling action logging and interaction with local models like Grog for debugging and AI tasks. It enhances problem-solving via local execution for privacy, and employs GPT-4 to streamline code writing in a REPL environment.
Jasper Schoormans - Fixing photos with Stable Diffusion.mp4 - I'm Jasper, a data scientist using AI to improve low-quality photos from a China trip. By applying stable diffusion and control nets, I've achieved consistent, aesthetic results. While challenges like altered faces remain, Confy-wide APIs offer promising opportunities.
Tomás Arribas - David vs Goliath the power of small models.mp4 - A specialized small model with 50 million parameters, trained on 30 million UK property addresses using the Roberta architecture, efficiently recognizes address components. It surpasses the Posta library by using fewer resources and better handling UK-specific formats, while maintaining competitive batch processing speed.
Fernando Rejón Barrera - GenAI in a prod first way.mp4 - Combine agents like RITFusion and high recall agents for relevant search results, evidence parsing, and comprehensive answers from multiple sources. Customize these agents with open-source components like Night Chain, keeping in mind that some platform tools remain proprietary.
Francisco Pereira & Sebastian Wehkamp - Choose Your Own Adventure GenAI game.mp4 - I am Francisco, a machine learning engineer at ML6, and I created an AI-driven "choose your own adventure" game during our Christmas projects week. This app uses GPT-4 for narrative elements and stable diffusion for visuals, delivering an engaging storytelling experience with imagery and sound.

Explore insights on Generative AI, AI development, and innovative AI applications.

Gray Blog

Join the inaugural AI Tinkerers meetup in Amsterdam, hosted by ML6, with food and drinks from Portkey. Share your AI demos, connect with practitioners, and dive into LLMs and generative AI. Space is limited, apply now for this exclusive event for AI builders and innovators!

AI Tinkerers Amsterdam - Inaugural Meetup - May 

Join us for the inaugural AI Tinkerers meetup in Amsterdam, hosted by ML6 🙌 and with food and drinks courtesy of Portkey 🍕

Following the popularity of the vibrant community of AI Tinkerers in London, Berlin, San Francisco, Seattle, New York and elsewhere, we’re thrilled to expand our reach and bring the magic to Amsterdam! This is your chance to showcase your latest AI demos, connect with fellow practitioners, and dive deep into the world of LLMs and generative AI.

Space is strictly limited and we will be screening attendance.

🔥 Apply now for an evening of hands-on demos and talks and connect with a community of active builders! Participate by presenting your own work, or just share ideas with like-minded people.

RSVP in the form above 👆
(location details will be revealed after RSVP confirmation)

AI Tinkerers is a meetup exclusively for practitioners with technical, machine learning, and entrepreneurial backgrounds who are actively building and working with foundation models, such as large language models (LLMs) and generative AI. If you’re deeply passionate about creating LLM-enabled applications, have hands-on experience in building such systems, and want to connect with like-minded people who share your level of commitment, then this group is the perfect fit for you. With AI Tinkerers meetings happening in many cities, we support a dedicated global community.

We are not just “AI enthusiasts”, we are AI Tinkerers. The core essence of AI Tinkerers is active collaboration around early-stage discovery and innovation, which requires a high degree of experimentation, vulnerability, openness to sharing challenges and learnings, and collaboration among individuals with a shared level of expertise. This unique environment allows us to embrace experimentation, foster an atmosphere of trust and innovation, and create world-changing solutions. 🌍

Ready to showcase your groundbreaking AI research, revolutionary AI use, or game-changing model? This is your moment to share with like-minded AI tinkerers!

We welcome talk and demos in the following categories:

⁠Applied AI or Gen AI (10-15 minutes)Technical AI Research (10-15 minutes)⁠Lightning Demos (5 minutes)

These will be short, casual talks, with heavy emphasis on technical showcasing. We encourage all attendees to demo what they’re working on.

Pitch decks are not allowed - we want to see your work in progress, no matter how messy!

Please note that capacity is limited, so we’ll prioritize those who are able to present a demo.

👉 Fill out the demo/speaker submission form here.

18:00 Doors open (⚠️ Please be on time!)18:30 Talks + Demo [1st block]19:10 Break19:20 Talks + Demo [2nd block]20:00 Networking21:00 Hit the pub? :)

If you are new to AI Tinkerers, please read the AI Tinkerers FAQ and join our Discord channel: https://discord.gg/8WsqnC6Hk3

PLEASE RSVP AT THE TOP OF THE PAGE ☝️ - space is limited.

After 15 years with Unity, ongoing frustrations over inefficiencies and tech limits exist despite AI advances. Apps help by automating tasks like transcription and invoice payments, simplifying communication and daily activities, but challenges and gaps persist.

I’m not a firefighter

I am Amir, founder of a Rotterdam-based startup, simplifying AI for developers since 2018. We've developed a plug-and-play system for seamless AI integration, focusing on personalization and context, unifying frameworks for efficient app creation without extensive tech expertise.

What the f**k is an AI-Native app

The LM terminal seamlessly integrates with the shell, enabling action logging and interaction with local models like Grog for debugging and AI tasks. It enhances problem-solving via local execution for privacy, and employs GPT-4 to streamline code writing in a REPL environment.

unicornsh A LLM in your (zsh) shell

A specialized small model with 50 million parameters, trained on 30 million UK property addresses using the Roberta architecture, efficiently recognizes address components. It surpasses the Posta library by using fewer resources and better handling UK-specific formats, while maintaining competitive batch processing speed.

David vs Goliath the power of small models

I am Francisco, a machine learning engineer at ML6, and I created an AI-driven "choose your own adventure" game during our Christmas projects week. This app uses GPT-4 for narrative elements and stable diffusion for visuals, delivering an engaging storytelling experience with imagery and sound.

Choose Your Own Adventure GenAI game

may-2024

What is AI Tinkerers?

Who is this for?

🎤 Demo / Talk Submissions

🕕 Event Schedule:

More information

Fine tuning LLMs for function calling

An AI photobooth prototype, designed for a wedding, uses stable diffusion 1.5 to randomly modify photos without props. It runs locally on a laptop with an RTX 3070, employing face and depth masking with control nets, enhanced by the Imagine AI Room library.

AI Photobooth

I'm Jasper, a data scientist using AI to improve low-quality photos from a China trip. By applying stable diffusion and control nets, I've achieved consistent, aesthetic results. While challenges like altered faces remain, Confy-wide APIs offer promising opportunities.

Fixing photos with Stable Diffusion

Combine agents like RITFusion and high recall agents for relevant search results, evidence parsing, and comprehensive answers from multiple sources. Customize these agents with open-source components like Night Chain, keeping in mind that some platform tools remain proprietary.

GenAI in a prod first way

june-2024

AI Tinkerers Amsterdam - June edition

AI Tinkerers returns to Amsterdam in September, hosted by Fiberplane and NP-Hard Ventures, with support from Groq. Join for AI demos, talks, and networking focused on LLMs and generative AI. Space is limited; apply to connect with builders in this exclusive practitioner meet-up.

AI Tinkerers Amsterdam - September edition

After a summer break, AI Tinkerers is back to Amsterdam!

Hosted by Fiberplane & NP-Hard Ventures and with food and drinks courtesy of Groq.

Showcase your latest AI demos, connect with fellow practitioners, and dive deep into the world of LLMs and generative AI with the best technical community in the Netherlands.

17:30 Doors open18:00 Demos (including 10 min break)19:30 Networking21:00 Hit the pub? :)

Groq is fast AI Inference. The Groq LPU™ AI Inference Technology delivers exceptional compute speed, quality, and energy efficiency. Groq, headquartered in Silicon Valley, provides cloud and on-prem inference at scale for AI applications. The LPU and related systems are designed, fabricated, and assembled in North America. Learn more and try Groq speed today at groq.com.

Fiberplane, the go-to tool for testing and debugging Hono.js APIs with full source code context and runtime.

NP-Hard Ventures, an early stage VC Investing in product-obsessed founders, creating the building blocks of tomorrow.

For AI builders, researchers and practitioners

Exploring unlabeled data classification with efficiency, I demonstrated using Spotify feedback by splitting it into requests, issues, and positives. Our app converts feedback into actionable insights through dynamic tags and automated classification, supporting product strategy.

Classifying Data with LLMs

Gorov's project uses AI to summarize YouTube videos, providing key insights and chapters. It's available as a Chrome plugin for content-heavy users. Aiming for high quality, Gorov is open to user feedback and collaboration for further development.

Summarise.live

I'm Kanish, in Amsterdam for work, and I've developed Whisper Clipboard after attending AI Tinkers events. It's a stack-based tool that enhances language models by mixing questions with context, making data entry easier.

Whisper Clipboard

Developed an app to convert interviews into reports, facing challenges with misgendering due to model limitations. Attempts to guess gender from transcripts led to errors, like suggesting HR candidates as firemen. Key lesson: separate prompts from example data to prevent errors.

As a computer engineer at Grok, we specialize in end-to-end tech stack management, from chip design to cloud API access. Developers can freely access our APIs via console.grok.com, and we offer tools like Scribe Wizard and Stockbot on GitHub.

ScribeWizard and StockBot powered by Groq

Stefan transitions from physics to computer vision in industry and police work, and develops a fitness app. Currently, the focus is on enhancing workplace productivity with scientifically-backed scheduled breaks and Slack integration for improved health and efficiency.

Mobi: The workday AI health coach

I'm Laudinas, a product engineer at Fireplane. Discover how an API client can automate request pre-filling, integrate request data, diagnose errors, and generate test cases using language models with a 100-line express-like API, Hano. It's open-source on GitHub and supports OpenTelemetry.