Friends of Dentro | October Update

Welcome to the very first edition of Friends of Dentro! 🎉

We set up this newsletter to keep you in the loop of what we’re up to at Dentro. Our work with clients, our own products and everything along the way. A wild mix of learnings, thoughts and things we’re working on. For anyone interested in AI, digital products and Dentro in general.

We believe this will evolve over time as we continue to learn and we plan on sending monthly-ish updates straight to your mailbox.

Thanks for being a part of it from the very start!

Topic of the Month:
Most companies don’t train AI models, they just use them

When talking to non-technical people, we often hear phrases like “we want to train a model,” “we’re unsure if we should build a model,” or even “let’s train a neural net.” It’s understandable why many think this way, but it’s also a bit off the mark. So, we’d like to shed some light on what options companies really have for using AI:

1. Use an API: The simplest option. Many AI models, both closed and open-source, are accessible via an API. OpenAI, Anthropic, Gemini – you name it. Just send a request and get the result back.
   Straight forward to use, but this alone doesn’t allow for usage with internal company data and you’ll need to be aware that you’re sending over data to those providers and choose the right provider, which can be an issue especially for European organisations.
   

2. Implement RAG (Retrieval Augmented Generation): This approach combines internal data with model outputs by vectorizing knowledge and blending it with generative AI.
   Sounds more complicated than it actually is. You can think of it like this: instead of sending your request straight to an LLM via its API, the RAG setup first searches your internal data, identifies relevant parts, and sends this over in the same prompt. This gives the AI model a chance to take your individual information into account as well when generating an answer (instead of only using what it knows by default).
   

3. Fine-tune an LLM or build a model from scratch: The former means tweaking a general AI model with additional data, the later means starting even a step earlier and building the whole model by one’s own.
   Both options require vast amounts of data. And both options are overkills for 99% of AI projects. It’s definitely nothing a company will or should start out with while being new to AI, but especially fine-tuning can be a viable route for more experienced organisations who want to take things a few levels further.

So most companies we work with don’t need to train new model, they need to use a model in combination with RAG. For a deeper dive into this topic, check out one of our blog posts from earlier this year: https://dentroai.com/implement-ai-in-a-company/

Tweet of the month

In light of all conversations going on about whether Europe is falling behind or not, that’s a interesting way to look at things. Many people in the comments below this tweet confirmed this impression and from our experience, we can’t argue much with it either.

We decided to put it to a test and start targeting US based clients in our marketing activities. With a more straight-forward approach compared to our approach in Europe, because if the hypothesis holds that US companies are quick to act, that’s what should get our feet into some doors.

We started building a dedicated landing page and will soon start reaching out to our target group in the US. More on that (and our results & learnings) in future editions of Friends of Dentro.

A peek into what else we’ve been up to in October

Building 🛠

• Built a suite of AI-powered customer service tools – a phone agent with VAPI, automated email responses through an Outlook plugin, and a responsive website chat that can be integrated on any webpage. Each tool pulls from internal company knowledge to deliver informed, tailored interactions, reducing the need for manual responses. Currently improving everything based on domain expert feedback and aiming for the user experience to feel as natural as possible.

• Completely restructured an AI chat app, which we’ve build some time ago for internal use by employees of a company. It’s now based on an AI agent infrastructure and works way smoother. The agent framework is responsible for using the correct tools, such as generating text, generating an image, handling a file, browsing the web, etc. Early numbers show increased usage since the update, which is definitely nice to see.

• Build our latest product NoteThisDown – An app that lets you transcribe handwritten notes with AI and store the results straight to a new page in your Notion account. It’s our very first products with a paywall (no freemium, just a 7 days free trial) and we launched on Nov, 1. More on the results and learnings in the next Friends of Dentro edition.

• Working on automatic offer generation for construction companies – Tested and iterated with real life examples of offer documents and pricing tables. We feed all information to AI and let it calculate the various positions within an offer to potentially save many hours of repetitive manual work.

• Building an AI agent who assists in writing blog posts – We have tried different blog post apps in the past and have never been happy with any of them. So we decided to give it a shot ourselves. Our approach is to lay out the different parts of a blog post first (title, outline, headings, images, sources, etc. etc.), bring them in an order that makes sense and let the AI agent iterate through them, only proceeding to the next step once the previous one has been completed successfully. First results were promising. It’s still work in progress, but we’re getting there.

Background activities 👨‍💻

• Set up this newsletter – You won’t believe how many newsletter platforms are out there 😅 We did a bit of testing and settled for Beehiiv for the time being.

• Tested a couple of new technologies:

  • PikaPod – easy hosting for open-source web apps.

  • GraphRAG – technique to generate knowledge graphs from raw text.

  • Umami – a web analytics tool.

  • Litlyx – another web analytics tool.

  • Plausible – the web analytics tool we finally went with for the foreseeable future.

  • Cloudflare and R2 buckets – cloud storage with a simplified approach to access stored objects.

  • Bolt.new – platform to build and launch full-fledged web apps with natural language (didn’t convince us yet, but this approach will be the future for sure).

• Wrote a whitepaper on AI integration in mid-sized companies that can be accessed here (in German): Link to whitepaper

• Experimented with latest image generation models, especially Flux and Ideogram. The results are pretty impressive!

Miscellaneous ✨

• Hit 100 subscribers on YouTube (standing at 141 at the time of writing).

• Got a very harsh email from a prospective client who replied to a project indication with just one word: “GAMEOVER!”. We decided against shaming him publicly (which frankly he would have deserved). Instead we sent him a few well-meant thoughts on professional behaviour, which at least made him apologize politely.

• Email exchange with LangChain CEO Harrison Chase – nice to see CEOs being this hands-on when it comes to their product.

• Tweet about Paul’s experience with learning about LangGraph went somewhat viral and reached 44k views, here’s the link if you want to check it out.

• Bought a couple of new domains related to indie hacking and AI agents. We believe both topics will rapidly increase in significance in the near future, so we want to make sure to have our hands on some good domains early on. Frankly, we are not entirely sure what to do with them for the moment, but we’ll figure that out along the way.

• Been thinking about different payment options for digital products: freemium, subscriptions, credit based, one-time payments,… there’s a lot of ways to do it! Our understanding of their pros and cons gets noticeably better and we’ll continue to experiment to see what works well.

And this wraps up the very first edition of Friends of Dentro. Thanks so much for being a part of it from the very start!

Questions, feedback, or anything else you’d like to tell us? We’re happy to read it all
– just hit ‘Reply’ on this email.

All the best,
Paul & Paul from Dentro

PS: You can share this newsletter with anyone who might be interested, just send them this link: https://friendsofdentro.beehiiv.com/subscribe