Open source startups are not a new concept: free and open source software have been around since the 90s and have been improving frequently on quality and delivery.

I’ve started talkd.ai as a chance of starting a new company, a new entrepreneurial journey where I would be leading a new approach on customer support and frequently asked questions (the original name that I had given to the project was FAQtor, but Avelino (the current Buser CTO and a long time friend) convinced me it was a lame name and talkd.ai would be better).

We’ve failed achieving our goal of becoming an open source startup, but yet our open source software is still here (and is here to stay as long as we have necessities around it).

Looking through the positive lenses, here are some outputs from this whole journey:

Getting Buser to test my solution

The beginning of this project is basically on a chat between me and Avelino, where I showed him a quick demo about an LLM answering some FAQ and being very accurate and helpful for the customers.

The project got started as a toy project, but getting production grade approval from a company that I admire was a ground shaking moment for me.

After this kickstart, Buser became an official sponsor of the project: sponsoring talented engineers time (shout out to Luan, Walison, Schumman, Erle, Bia and Caixeta) to dedicate to this project.

In this operation, we were able to fully respond over 3 thousand support tickets in the operation start month.

Being GitHub Accelerated was incredible

Seriously, Github Accelerator was incredible, it gave me a lot of confidence and feeling that we had a very solid solution for a suboptimized market.

Gregg and Kevin (program managers at Github Accelerator who lead us and helped us achieve our wildest open source dreams) were super friendly and helped us a lot during the whole accelerator, they brought all of their know how and talent from previous business/investments.

Being part of such an amazing cohort made me realize that open source can bring you to collab with multiple awesome people working on different technologies. Namee from LLMWare is using amazing techniques to optimize SLMs on CPUs; Daniel and Michael from Unsloth are rocking grounds in the Valley; Diego is improving the VR scene with A-Frame and AI integrations; Jurgen and Andreas are rocking and hacking lots of servers with hackingbuddygpt; Alicia is doing some amazing work on InfoSec information; Michael from Langdrive is working on amazing mortgage brokerage tech; Tim and his sweet open source work on OpenWebUI is amazing and Marimo is… well, Marimo! Akshay and Myles are excellent engineers.

This whole connection and group setting enabled me to get a nice overview on the international landscape and how far behind we are in Brazil and how we can advance the knowledge locally.

Github Universe connected everything together

Me on the entrance to GitHub Universe

Being brief here, Github Universe enabled me to connect all of the dots from the past to the moment.

It made me understand why I value open source so much and why I do still believe that this is the best model for a good technological future.

I was able to get involved in multiple awesome agentic workflows

I worked on 2 politician bots fully automated with AI; Buser’s sales and support agent and other many smaller projects.

The technology is here to stay, more people are embracing it and more and more solutions are using groundbreaking scientific knowledge to better create human-machine interfaces.

But now, the not so obvious or pretty part… the market is rough and technology is initial

Who was thinking about using LLMs for customer support? ~no one~ Everyone was looking at the same possible use cases as I was.

The solution for the problem is quite simple: you just define a very good FAQ content, deploy it to your vector database and improve the prompt in order to make it use the voice tone and brand aspects you want to.

WRONG! There are lots of usecases where, having a misleading prompt or when the LLM hallucinates, a major customer support issue can popup without the proper guard rails.

Chevrolet even lost some money selling a Tahoe (a car we don’t have in Brazil) for a single dollar.

Competition is very talented

Even though every single person doesn’t have all of the knowledge of the world, there are thought leaders that can create a new wave of concepts using a single simple idea.

Cloudhumans (Brazilian company) are basically tackling the same problem on the commercial side, and they are doing amazing tech with ClaudIA.

Advolve is working on top notch Martech AI, fully automating creative and strategic marketing workflows.

Those are just two examples, major players are already implementing the LLM layer onto their support software, which, as they already have customers and incoming tickets for those customers, they will have the lead of the market.

I didn’t find a Product Market Fit

Honestly, I did search for a market that would receive our technology and know how, but I wasn’t able to find a good enough market that would justify spending the next 20 years working on it.

I do care about how AI can leverage our knowledge and empower better customer support, but I really think the underlying technology is cooler than the application that I found.

Looking backwards, the perspective was different some months ago. The market was still very young and not many models were available, but right now we are looking through a very accelerated landscape.

TLDR

If you don’t want to read the whole article, here is a brief summary of what I shared here:

1) Avelino is an excellent partner and open source contributor

2) Buser is an amazing company (now I’m working for them and maintaining our talkd.ai instance internally)

3) Github Accelerator and Universe changed my life, gave me awesome friends I’ll never forget and connections that I have never thought I would make

4) The Customer Support technologies using LLMs are getting widespread, making the market rough and very competitive (race to the bottom is not my type of strategy)

5) I didnt find a quite good PMF with my experience and knowledge

Is talkd.ai dying?

No! It’s not! We are still up, running and proud to be open source.

But the business talkd.ai is not worth investing money or time into making it a VC backable business, that would require more niche tactics and specific knowledge that is not under my watch and scope right now.