To be very honest, developing Scott is the most ridiculous project I have participated during my entire career, but yet, I count it as one of my proudest achievements.

The article below focuses mainly on the overall journey of how Scott was developed. We will release the technical adventures in the future if you are interested.

All In The Golden Afternoon

It was during the end of 2022 when my company, Intelllex, had finished SEO and site optimisation for Finreg.sg, and was still figuring out the use case, when we decided to explore lead generation as a business model.

To give everyone a bit of background, Finreg.sg is a regulatory search engine providing bite-sized financial regulations content. Personally, I think we’ve done a great job even though it didn’t involve cutting edge technology. At that point in time, no one believed nor realised that the world will be changed one week after that.

Down The Rabbit Hole

When ChatGPT comes out, engineers are shocked and excited about it and we started to spend some time doing research after work as a hobby. Not long after that, there were daily news and information related to ChatGPT, Generative AI, and LLM. Realising the potential, our company started to explore harnessing this new technology even though we have zero knowledge nor yet to learn how to use GPT technology in building a product. There were endless discussions on the direction/approach and potential products we would like to create, but the biggest challenge was that we were hampered by our limited knowledge of GPT.

To resolve this situation, my Data Science colleague researched thoroughly and tried his best to explain the mechanism in layman terms to the other colleagues. This behaviour has proven to be the most essential part of the development process. It was with this rudimentary understanding of how ChatGPT and LLM work, and the trust among the various stakeholders, especially the CEO and CTO, that this project will succeed.

The One-day Prototype

At the start of Q1 in 2023, we were feeling stressed since some applications had been released to the industry, claiming to be built on or used by ChatGPT technology.

We started to suffer since then. It’s no joke dealing with the novel technology – everything was still determined and unproved – including the technology stability, product, design and market fit.

Our first try was to build a chatbot that can answer legal questions using GPT technology which I received orders to build the prototype in one day. Yes, you heard that right. I have been asked to complete a working prototype in one day. the end result was a model offering syntactically correct but bogus answers, which in my view, has failed the expectations.

However, looking back on the way we had developed the prototype, I think this was a blessing. We realised GPT will generate unrealistic results before others found out. We also did not try to hide or escape the limitations; we faced it as a professional. We learn and understand that technology still has limitations in the very early stage.

Back To Start

After Lunar New Year. We started brainstorming to figure out a real-world use case of GPT, which was the second right thing we’ve done.

In my opinion, the value of AI-related technology is that it opens new possibilities to reduce the cost of doing things. For example, before GPT, researching still takes a lot of time even though we have powerful search engines such as Google. People still have to read through the various search results to find the answers. But now, GPT allows people to obtain the answer directly after asking a question – this could save massive costs in any industry in the long run.

Based on this principle, we started asking ourselves, what is the first thing we must build? The answer is well, surprisingly, still searching. The difference between searching using GPT compared to the traditional way (such as text-based, boolean-based, or graphic-based) is that using natural language as a query can provide more accurate results. This way, people don’t require high search skills to get the desired answer.

Red Pills Or Blue pills

March 2023, I will never forget March 2023, for it represents one of the biggest failures in my life. We had built a demo an AIWM event, but the demo was broken during the presentation. The important lesson we learned was that the product had to be tested continuously and seriously, even if it’s just an simple 10 minutes’ demonstration. Testing is such an important component in the entire development cycle – it’s the risk analysis and management for the whole company – regardless of company size – from startups to FAANG.

The product stability must depend on the test results and not on guesswork or feeling. People who don’t care about the result can choose the blue pill, preferring to live inside a simulation, but not me. I care, so I will write the test before releasing anything for the rest of my life. Welcome to the real world.

Solving Problems

April 2023, Another critical question to answer when building a product is: What goal do we want to achieve? We can start by discussing the law industry. Law is one of the oldest industries in human history, but it still has a lot of issues that need to be solved. We might have various tools to facilitate a lawyer in modern age, but some of the solutions are still costly and time-consuming. We want to resolve this issue by optimising the workflow for lawyers worldwide and this is the reason why we created Scott. The product direction and features we want to build become more coherent after clarifying with the team on the product vision. We have a better idea of features that can alleviate lawyers’ pains.

Building Trust By Showing The Truth

Law, like most knowledge-intensive industries, must be accurate and based on facts. We have seen many articles showing people paying the penalties who used GPT generated results without verifying. The impact is that people will not trust AI if technology advancements don’t focus on producing accurate results. Our answer to the market is to provide accurate, trustworthy, and results with supporting authorities. This is where we put the most weights next in May 2023.

Finally, Meet Scott

May 2023, Scott was released the public on May 30, 2023. We are all excited to show the world how Scott can help lawyers solve their problems so that they can have better work experience and enjoy life better.

If you interesting about how we doing, leave your comment below, I love to help!