An old colleague has launched a new meetup in the Cheltenham area, titled the Massive Graph Analytics South West Meetup. I attended the first in-person meetup, kindly hosted by Ripjar.

If it seems niche to you - Gloucestershire has an unusually high population of graph data engineers and data scientists. 15 people isn't bad for a first meetup on any topic. Tech meetups are just informal get-togethers with no ceremony. The drill is typically pizza, beer, some presentations, and a chance to just discuss a topic with people with nothing more in common than a passing curiosity.

It was great to meet up with old colleagues Simon and Rob who now work at RipJar. Simon gave a really insightful presentation on Ripjar's graph analytics embedded in the Risk platform.

My contribution was a presentation on Graph RAG which is a technique for integrating knowledge graphs with AI chat services. This had jokingly been applied to the topic of spacecraft safety engineering - although people may not trust AI for rocket science, the topic area provides a useful data environment to test deep reasoning algorithms.

Tradtional RAG techniques for pushing documents into AI services struggle when applying a task that requires touching many different and unrelated pieces of information. Graph RAG is a fascinating technique for adding knowledge to AI-enabled services in a way that supports much deeper knowledge for complex reasoning tasks. It's interesting that commercial LLMs are equipped to deal with knowledge graphs, and yet the idea of using that nacent capability to support introducing knowledge into AI pipelines is incredibly recent. I gave a demo of using Graph RAG to reason about failures in the space shuttle program, which showed the LLM's ability to reason with unrelated chains of events and propose candidate cause-and-effect hypotheses.

Talent

Changing the subject, earlier in the week I had been present for another panel discussion where startup founders were talking about the difficulty in obtaining the skills needed to grow their businesses. It's a daunting business launching a startup as a co-founder. Even if you have some experience of recruiting talent in a business, it's a very different story working in a business, with limited or sporadic cashflow. Unless you have very understanding investors, tapping into the top talent is going to be problematic.

This took me back to my time as a co-founder (CTO) of a cybersecurity detection business, living and breathing the problem of talent acquisition. While good talent is a scarcity, we did OK. Interviewing people for a startup sometimes feels like a reverse-interview where you have to persuade the interviewee that your startup isn't a completely insane idea.

For some jobs we had to be flexible, and consider those who wouldn't be a perfect match for the job description if we were working in a big company. The trick is to consider flexbility, adaptibility and ability to learn. These are all useful traits to have for anybody working in a startup, anyway. It turns out that aerospace data scientists can learn cybersecurity data science and physics researchers can be cybersecurity researchers. There are lots of ways to connect with people who might want to be in your business. Tech events like meetups are worth going to if you're recruiting.

It's useful to remember that in general, most people do a bad job of recognising talent. People don't want to take risks in recruitment, and that translates into 'safe' decisions, and recruiting people who conform to a template or mould, ignoring potential. If you're a startup founder struggling to acquire talent, remember this, and consider that you may be missing opportunities.

Lack of diversity in recruitment is an equality problem, but also has major economic impact.

I'm writing this on International Women's Day, and just read a blog post which reported that women-led businesses receive just 3.5% of equity investment, which is a colossal failure to recognise talent.