I’m a Luminance veteran – I joined the company six years ago as a software engineer after completing my PhD in Quantum Computing at the University of Oxford. Now, I lead the company’s research efforts and split my time between managing the R&D team and conducting my own research into the latest machine learning techniques, mostly around building Large Language Models (LLMs for short) and their applications.

I find it a lot easier to collaborate in-person so typically I go into our Cambridge office from Monday to Thursday. I can normally be found in the ‘Lumi Lab’ – a new room for the R&D team that is full of whiteboards so we can hash out ideas together.

Here is what a typical day looks like for me.

9:30am

I arrive at the office and check the results of some code that I ran overnight. Today, I’m looking over a new version of ‘Ask Lumi’, our brand new, AI-powered chatbot which allows lawyers to ask questions about their contract and receive instant results, or even redraft clauses, all from within Microsoft Word. I was tweaking the code to alter the output of Ask Lumi and see how that affects certain categories of questions. I send the results off to our brilliant Legal Analyst team who check the responses against their gold standard answers, ensuring the utmost legal rigour.

10:00am

We have our morning meeting led by our CTO, Simon, and Director of AI and Operations, Graham, where we discuss our priorities for the day as a team and share the results of any code we ran the night before. Although I had some coding experience prior to joining Luminance, I had been worried that software development would be a lot of plumbing together of other people’s tools. Luckily, that hasn’t been the case at Luminance as we’re always working at the cutting edge and developing new techniques to solve interesting business challenges!

10:30am

After the meeting I spend some time glancing through the latest academic papers my team has identified as having the potential to improve our product and pick one as the topic for our reading group next week. It is really important that we stay on top of the latest research and understand any new technologies which could have applications within our products. While generative AI is getting a lot of attention right now, we’re excited to be able to work on the next stage: what we like to call ‘specialist’ AI. In industries like the law where specialist, domain knowledge is required, this type of AI is crucial, so it’s exciting to be working at the forefront of these developments.

11:30am

I’ve received the Legal Analysts’ feedback on the code I ran last night. There were a few cases where the prompts fed into Ask Lumi haven’t worked as well as I’d hoped, so I investigate these now. I’m trying to identify any patterns in the questions the model struggled with most. In this case I spot that a number of them mention relative dates, e.g. ‘How many of our NDAs will lapse within the next 30 days?’, and that Ask Lumi responded to say it doesn’t know today’s date. I modify the LLM’s system prompt to include today’s date and try a few versions of the prompt with different date formats until I am happy with the response I get.

12:30pm

Lunch! There is a new vegan place down the street from the office called Urban Butterfly. I pick up some food and eat it in the nearby park.

1:30pm

We have recently started a hiring push, so my next task is to interview one of the candidates. We hire people with all different skillsets, degrees, and backgrounds, but they all have one key thing in common: excellent problem-solving skills. Two days before the interview, we send them a paper to read. For the first half of the interview, myself and Julio, Head of R&D, get to know the candidate and talk about Luminance. For the second half, my colleague Kuan in the R&D team will join the call and we’ll ask the candidate questions about the paper.

2:30pm

I dedicate two hours in the afternoon to my ongoing research project – I can’t say what it is right now, but it's very exciting!

4:30pm

Reading group time. Around seven of us in the R&D team gather in a meeting room to discuss last weeks’ suggested paper. These can be tough to wrap our heads around, but together with the aid of whiteboards and videos, we make steady progress. We make sure to write up a short summary and create Research Proposal tickets for any ideas we think could improve existing functionality or add new features to help our clients.

5:30pm

I’m setting up a new Large Language Model I’m experimenting with to fine-tune on our legal datasets overnight and then answer emails that have come through in the afternoon. We’ve recently started a new activity within the R&D team – all members update a communal list of daily tasks to clarify what has been completed and come up with new tasks for the next day. This is a great way to track our progress as a team, share updates in tomorrow morning’s meeting and make sure any achievements are recognised!

6:00pm

It’s the end of the day and I’m going bouldering with a group from the office. Whether it's bouldering, the pub, or another activity, there is always something going on post-work with the Luminance lot!

Interested in finding out more about working at Luminance? Visit our careers page.