Meet Jamie, Skillmint's Technical Co-Founder

Hi. Hello. Welcome!

I'm Jamie, the technical co-founder of Skillmint and this is my first post. The last 2 and a half months have been a non-stop speed run of building Skillmint, an AI-led product to help people hire better, faster and with less bias. All this in between life and raising a toddler. Over that time, I have had to make choices in how to build the best product possible, what to build vs what to buy, what frameworks to use, what features to include and exclude, and so on. This post is a barely coherent, barely structured, written by an engineer's account of those big decisions. Hopefully you find something useful!

The Big Unspoken, Unwritten Deadline

I have heard, more than once, AI being described as a Gold Rush. I believe it's a little more transformative than that, but I think they do share certain characteristics. The introduction and evolution of LLMs has opened the possibilities of solving problems in nearly every area where previously the technology just couldn't meet the need. Currently in most sectors and markets, the landscape is so new that no one dominates yet, companies are showing up to this unspoilt landscape with amazing things that provide new solutions to problems that used to be intractable and people are rightly wowed.

However, this is not going to last forever. Products like Cursor or n8n (or even Skillmint!) allow new entrants to build and hire at a searing pace, and there is a very real possibility that an opportunity today won't be there in a couple of months. With this in mind, I had a very big unspoken, unwritten deadline to get Skillmint done. Something which was extremely useful in providing focus, motivation and a lens for making decisions.

Build vs Buy - The Eternal Debate

This is always contentious. Having spent a decade working in data, I have lost count of the times spent in the pub with friends in similar roles bemoaning the Drag-and-Drop merchants who have somehow made their way to Heads and Directors of data positions without the ability to write a single line of Python or SQL.

Because of that, I have been pretty heavily biased towards building. But at Skillmint we are all about reducing bias, and this gave me the opportunity to rethink things. In the last two and a bit months, I have discovered that there is actually a sweet spot where you can have professional pride in your work and pay for things that enable you to get to where you want to go both better and quicker. For me, it was made possible by a set of fundamentals:

• The product has to be ready in less than 3 months

• The product needs to have the following 2 features, Screening and our soon-to-be-launched second feature

• The product has to be secure and reliable

• No vendor lock in

• The product needs to be profitable

• There is only me

These 6 unalterable fundamentals provide a very useful checklist to base a decision off of. For example, when building Skillmint Screening, the first iteration was an attempt at fully creating the video call experience with transcription, recording, tab switching detection etc. by myself. It was close to unmaintainable and didn't work on half the available browsers or on mobile. 

Then I discovered LiveKit. As a teacher in a previous life, I was trained not to reinvent the wheel. LiveKit and its agent framework allowed me to allocate my time to novel problems whilst it handled the calls infrastructure fantastically well. LiveKit is well maintained, within our cost requirements and actually allows us to be less locked in with LLM providers, giving us the ability to test out different conversation, turn detection and TTS models in minutes. At the same time, it's not something that can just be dragged and dropped together, a big tick for professional pride!

Build v Free - Open Source Frameworks

There are A LOT of open source LLM frameworks these days, sometimes to a point where looking for one that fits your needs feels like it would take longer than building one from scratch. Sadly, quite a few of these frameworks aren't great, I was in the middle of using one which broke when Google made a change to how they work with bots, and remained broken for weeks without any updates, before I eventually gave up.

That being said, there are some great, professional open source frameworks emerging. We use LangGraph and MCP extensively for our evaluation agents. The ability to spin up complex, multi-agent architectures in an hour or two is, without overstating it, the difference between Skillmint launching now or in 6 months, if at all. Again, it also allows us to test and use multiple models, add fallbacks in case of failures and easily link to monitoring and evaluation.

There are some great tutorials for making your own supervisor-orchestrated, multi-agent services. I recommend this one as a starting point: https://langchain-ai.github.io/langgraph/tutorials/multi_agent/agent_supervisor/#2-create-supervisor-with-langgraph-supervisor

For me, the most exciting and terrifying thing about AI and LLMs at the moment is the unrelenting pace of advancement. LangGraph Supervisor, which is quite important for parts of Skillmint, was released in February this year. If we had started a couple of months earlier, things may have been very different, and undoubtedly, something new and even better will be along before you know it.

The Future

As terrifying as the pace of advancement and the fear of keeping up might be, the outcome is still exciting. Solutions to problems that existed outside of the realms of possibility are now possible. 

At Skillmint, we are solving the problems inherent in the recruitment process. Everyone knows what it's like to be perfect for a role only to be ghosted after applying. Every hiring manager and recruiter knows what it is like to receive so many applications, real and AI generated, that it is impossible to give more than a minute or two to each to read and compare them consistently, potentially missing the ideal person. Personally, I'm excited about causing the long overdue death of the CV and Cover Letter!

If you are too, get in touch for a demo: https://www.skillmint.ai/book-a-demo