Sign in

Andreas

@andnig.bsky.social
174 followers 672 following 83 posts

Writing about Data & AI applications

PostsRepliesMedia
Andreas @andnig.bsky.social · 30/07/2025
We built an AI that: → Connects to your inbox and loads invoices, orders, etc. → Auto-extracts data from text + attachments → Shows everything for human review → Exports, after approval, to your CRM, etc. This saves us so much time. Built with @vercel.com AI SDK.
010
Andreas @andnig.bsky.social · 16/07/2025
There is not a single invoicing app for freelancers out there, offering these things: - no subscription, just pay per invoice - beautiful, professional invoices - just let me time-track and invoice. No bloat So I built my own. I don't know, is this for someone else too? Using @shadcn.com
020
Andreas @andnig.bsky.social · 24/06/2025
Just crazy. Anyone saying you can vibe-code yourself to a working SaaS is just lying. Just created this monster. I did not ask for this and used the Opus 4! AI is an awesome tool, but you are making a fool of yourself if you think you can autonomously AI-code a complete app.
010
Andreas @andnig.bsky.social · 01/05/2025
I can't comprehend how the US - undisputed world leader in almost any area - simply threw all that away in less than 100 days, making it necessary for companies to basically distance themselves from the US. 200 years of hard work, thrown away in 100 days. blogs.microsoft.com/on-the-issue...
blogs.microsoft.com
Microsoft announces new European digital commitments - Microsoft On the Issues
As a multinational company, Microsoft believes in trans-Atlantic ties that promote mutual economic growth and prosperity.
000
Andreas @andnig.bsky.social · 17/03/2025
🚀 Excited to share my latest blog on building AI agents with smolagents - a refreshingly minimal library for ai agents, created by @hf.co www.pondhouse-data.com/blog/smolage...
pondhouse-data.com
Building Powerful AI Agents with smolagents: A Minimalist Approach
Discover how smolagents provides a lightweight, flexible framework for building AI agents without the complexity of larger solutions. Learn how to create powerful agents with minimal code and maximum ...
000
Andreas @andnig.bsky.social · 06/01/2025
I failed as a SaaS founder, so I created a boilerplate for self-hosting applications on a VPS vps.pondhouse-data.com
vps.pondhouse-data.com
VPS Ready Next.js - The Minimal Self-Hosting Boilerplate
Production-ready Next.js boilerplate for cost-efficient VPS self-hosting
040
Andreas @andnig.bsky.social · 16/12/2024
In this step-by-step tutorial, we break down: - Setting up Azure AI Search for RAG - Implementing efficient data indexing and retrieval - Integrating with Azure OpenAI for enhanced responses - Best practices for schema design and system optimization www.pondhouse-data.com/blog/rag-wit...
pondhouse-data.com
Step-by-Step Guide: Building a RAG System using Azure AI Search
Build a RAG system with Azure AI Search: This step-by-step guide shows you how to integrate retrieval-augmented generation with Azure AI Search. Learn how to build more accurate and context-aware AI a...
010
Andreas @andnig.bsky.social · 11/12/2024
Google just released Gemnini 2.0 Flash and it's amazing. One of the best if not the best model so far. We successfully bullied Google in creating the best AI model. 😅
010
Andreas @andnig.bsky.social · 10/12/2024
Great post by Amos Gyamfi about Multi-Agent application platforms. The post outlines, what multi-agent systems are, why you want them and then introduces the following tools to help with that: - Langraph (by LangChain) - AutoGen - CrewAI - Swarm (by OpenAI) - phidata medium.com/@amosgyamfi/...
medium.com
Best 5 Frameworks To Build Multi-Agent AI Applications
This article aims to help you build AI agents powered by memory, knowledgebase, tools, and reasoning and chat with them using the command…
160
Reposted by Andreas
camwing @camw.ing · 05/12/2024
In 2025 (less than 30 days away), PayPal will start selling your transaction history for targeted advertising. I very highly recommend logging into your account and going to Settings > Data & Privacy > Personalized Shopping. If you're reading this, turn that off RIGHT NOW before you forget.
A screenshot of PayPal's terms of service. Big wall of legalese text, which I will be pasting in below. You don't need to read it, here's the important part: 

"The key update to the Privacy Statement explains how we will share information with merchants to personalize your shopping experience and recommend our services to you. Personal information we disclose includes, for example, products, preferences, sizes, and styles we think you’ll like. Information gathered about you after the effective date of our updated Privacy Statement, November 27, 2024, will be shared with participating stores where you shop."


Here's the whole paragraph: 

We are updating our Privacy Statement to explain how, starting early Summer 2025, we will share information to help improve your shopping experience and make it more personalized for you. The key update to the Privacy Statement explains how we will share information with merchants to personalize your shopping experience and recommend our services to you. Personal information we disclose includes, for example, products, preferences, sizes, and styles we think you’ll like. Information gathered about you after the effective date of our updated Privacy Statement, November 27, 2024, will be shared with participating stores where you shop, unless you live in California, North Dakota, or Vermont. For PayPal customers in California, North Dakota, or Vermont, we’ll only share your information with those merchants if you tell us to do so. No matter where you live, you’ll always be able to exercise your right to opt out of this data sharing by updating your preference settings in your account under “Data and Privacy.”
366110598254
Andreas @andnig.bsky.social · 05/12/2024
What do we all make of this? Is this good? Bad? For society I mean
010
Andreas @andnig.bsky.social · 04/12/2024
The worst thing for a company is, if Managers have too little work on their own plate. They'll annoy anyone else and reduce overall productivity in a company.
000
Andreas @andnig.bsky.social · 03/12/2024
Folks, if you use LLMs in Azure, deactivate these god-awful content filters. They are absolutely useless and slow down your application www.pondhouse-data.com/blog/azure-a...
pondhouse-data.com
Azure OpenAI Content Filters: The Good, The Bad, and The Workarounds
Azure OpenAI is one of the most used AI platforms for enterprise users. The Azure environments is relatively well-trusted and provides access to all OpenAI LLM models. By default, Azure offers 'Conten...
020
Reposted by Andreas
Dr. Damien P. Williams, dread portent down from a mountain cave @wolven.blacksky.app · 30/11/2024
Please stop uncritically sharing "AI" generated bullshit. You're doing so much harm to yourselves and our overarching ability to create and certify knowledge and understanding as a society. I know that sounds hyperbolic, but it's honestly the stakes here. I'm literally begging you.
15730116
Andreas @andnig.bsky.social · 29/11/2024
Friend consults a company in established industry. The comp had no growth in the last 6 yrs. They introduced a new product which is a slight variation of an existing, 10y old one. A new manager started - predicting 35% total revenue growth yoy each year for the next 5 yrs. What does my friend do?
010
Andreas @andnig.bsky.social · 29/11/2024
This is what I'm talking about. If we say we don't want to use bsky data externally - I'm in. Privacy is great. But if we allow it, we can't shitstorm if someone we don't like uses the data according to our own rules. For completeness & related: Huggingface is much more than CO2 jugging LLMs!
021
Reposted by Andreas
Adriaan @adriaandotcom.bsky.social · 23/11/2024
Created a Bootstrapped founder Starter Pack. People want to add interesting people fast. Hit add all if you want to follow interesting founders. Want in? DM go.bsky.app/E4f3eAx
12183
Andreas @andnig.bsky.social · 28/11/2024
currently learning about espresso. Super interesting. Can't wait to finally make my own, delicious, brown goodness. I found this guy here to be not only super helpful, but very very relaxing. Give it a shot (pun intended) www.youtube.com/watch?v=aTFs...
youtube.com
Understanding Espresso - Dose (Episode #1)
YouTube video by James Hoffmann
010
Andreas @andnig.bsky.social · 28/11/2024
I'm astonished how many people on here literally hate "AI" - as in the technology. Not LLMs, but AI. As if almost any of the advancements made in the last 15 years were not assisted by any form of AI... Sad to see this blind hatred.
140
Andreas @andnig.bsky.social · 27/11/2024
What I love so much about the OnlyPosts Feed here: You reload the page and you still have the same feed. It doesn't force you into this weird and addictive feed-reload just be reloading the page.
000
Andreas @andnig.bsky.social · 26/11/2024
Highly interesting post about how to get AI to create meaningful reports. Many insights also for non-report use cases medium.com/firebird-tec...
medium.com
How to make more reliable reports using AI — A Technical Guide
Technical guide, sharing my experience of working with AI
010
Reposted by Andreas
Jeremiah Lowin @jlowin.dev · 25/11/2024
Interviewer: what's your greatest strength? Me: i have exceptional hindsight Interviewer: that doesn't really help us Me: yes, i see that now
11787134
Andreas @andnig.bsky.social · 26/11/2024
One of the first starterpacks I used. Great people. The fun is back in social media.
010
Andreas @andnig.bsky.social · 25/11/2024
Nice work @anthropic.com . They just released a proposal for an open standard for interfacing data sources. This is something we need. I hope it catches on. www.anthropic.com/news/model-c...
anthropic.com
Introducing the Model Context Protocol
The Model Context Protocol (MCP) is an open standard for connecting AI assistants to the systems where data lives, including content repositories, business tools, and development environments. Its aim...
011
Reposted by Andreas
Simon Willison @simonwillison.net · 25/11/2024
Anthropic released an interesting thing today: an attempt at a standard protocol for LLM tools to talk to services that provide tools and extra context to be used other the models modelcontextprotocol.io
modelcontextprotocol.io
Introduction - Model Context Protocol
Get started with the Model Context Protocol (MCP)
1837652
Andreas @andnig.bsky.social · 25/11/2024
Many AI applications require realtime info from the web. Therefore we need to crawl websites quickly and reliably. If you require a self-hosted crawling solution with great DX, crawl4AI might be the tool you're looking for. Fast, and has a nice feature set. www.pondhouse-data.com/blog/webcraw...
pondhouse-data.com
Crawl4AI Tutorial: Build a Powerful Web Crawler for AI Applications Using Docker
Learn how to set up and use Crawl4AI's web scraping capabilities using Docker. This step-by-step tutorial shows you how to set up, configure, and deploy your first AI-powered web crawler in minutes. F...
000
Andreas @andnig.bsky.social · 25/11/2024
Currently working on validating all these AI pair programmers. Holy cow, Aider is by far the best. Not cursor, not bolt, not v0. Aider is so incredibly good. Why is there so much hype about the others, but not Aider?
010
Andreas @andnig.bsky.social · 24/11/2024
Small bsky hack: The "OnlyPosts" Feed is so good. Very focused dopamine kicks without engagement bait. I love it.
000
Andreas @andnig.bsky.social · 24/11/2024
Can anyone here give me some tips for @bolt.new ? It's good for bootstrapping a project but than it is very stubborn when it comes to changing things. And also I don't get the pricing. 20$ for 10M tokens. 10M tokens per month? This is kinda little, isn't it? But I think I just miss some things.
000
Andreas @andnig.bsky.social · 24/11/2024
How many starter packs is too many? Kinda addictive to follow all these amazing people. 😅
000
Andreas @andnig.bsky.social · 20/11/2024
Some time ago we made an intro article about how to automatically verify the performance of models using promptflow and Microsoft Azure AI Studio. I think this is quite a nice workflow to ensure your models still behave as you like after an upgrade. www.pondhouse-data.com/blog/azure-a...
pondhouse-data.com
Azure AI Studio: How to evaluate and upgrade your models, using the Prompt Flow SDK
Learn how to effectively evaluate and upgrade your AI models using Azure AI Studio and Prompt Flow, making sure that the new model does not introduce any regressions.
000
Andreas @andnig.bsky.social · 12/11/2024
This is kinda crazy: You can create embeddings which conserve context across chunks. Meaning, with RAG, you don't lose necessary context when chunking. www.pondhouse-data.com/blog/advance... #RAG
pondhouse-data.com
Late Chunking: Improving RAG Performance with Context-Aware Embeddings
Searching for the right documents and chunks is the most important aspect of a RAG application. However, different chunking methods come with different drawbacks. Using long-context embedding models, ...
000
Andreas @andnig.bsky.social · 06/11/2024
Hello world. Trying to create high-quality human-generated content about how to use AI for the common good. One of our latest posts: How to use DSPy, a great framework by Stanford University to automatically create better prompts. www.pondhouse-data.com/blog/dspy-bu...
pondhouse-data.com
DSPy: Build Better AI Systems with Automated Prompt Optimization
Learn how to use DSPy to automate and optimize your LLM prompts. This step-by-step tutorial shows you how to build more reliable AI systems without manual prompt engineering and serves as introduction...
020