Ever since announcing Koodoo-LLM's success in passing the CeMAP exam, I've been frequently asked about our reasons and objectives.

I firmly believed in GenAI's potential and was eager to explore its practical benefits for businesses, and that drove our attempt to deploy AI in our mortgage work. 🧭

Our initial attempts with general AI models in the mortgage sector showed us that while they’re good at basic interactions, they often struggle with the more specialised aspects of mortgages. Despite ChatGPT's ability to conduct engaging conversations, it struggled with the nuanced, context-specific topics and often hallucinated.😶🌫️

Think of it like hiring a customer service assistant for a mortgage firm; you’d expect them to be relatively fluent in the language of mortgages, familiar with terms like 'overpayment allowances', 'joint borrower sole proprietor mortgages', and 'minimum cladding requirements'. If they are not, they need to be trained by the firm before serving customers. This is similar to how we approached training our LLM – getting it up to speed with the specific language and details of the mortgage world.🎓

💻 To do this, we fine-tuned a generalist AI model and transformed it into a mortgage-specialised model - the Koodoo LLM.

Fine-tuning teaches a generalist LLM like ChatGPT or Gemini with specific data and scenarios. This equips AI with deep knowledge of a particular domain, enabling it to grasp not just the vocabulary but also the subtle aspects of the context. This is a step up from zero or multi-shot prompt engineering, as it involves modifying the model's fundamental structure, thereby expanding its range of practical applications.

🤨 Wondering if fine-tuning AI models is necessary for your business?

In my experience, applying AI in your business involves three crucial steps:

1. Identifying high-priority use cases,

2. Testing your ideas with basic prototypes (MVP),

3. Optimising for cost and accuracy across a wide range of scenarios.

Starting with general models and engaging in prompt engineering can effectively navigate the first two steps, achieving about 70%-80% accuracy.👌

But for full deployment, especially in fields where precision and reliability are highly necessary, fine-tuning becomes indispensable.

Fine-tuning allows for adaptation to specific business or industry contexts. In our case, it enabled the model to converse with mortgage-specific terminology, akin to a trained advisor. This approach is equally valuable in other sectors like insurance or banking.

🤠 The good news is, fine-tuning isn’t as daunting or costly as it might seem.

Thanks to the availability of foundational AI models and open-source technologies, fine-tuning is more about customization than starting from scratch. It significantly reduces the resources and data required to train a specialised model.

For Koodoo’s LLM to pass the CeMAP exam, it only took us a couple of weeks and a few hundred data points through the whole process. In fact, we discovered that using thousands rather than hundreds of data points worsened the model's performance!

Therefore, the time and data investment for fine-tuning is quite feasible and manageable.

🧑 The most critical component here is having the right talent.

To craft a customised AI solution that meets your specific needs, you will at least need a skilled data scientist knowledgeable in various aspects of GenAI and a well-prepared dataset - which currently isn’t available in-house for many enterprises.

Of course, not every company needs to become a GenAI powerhouse. But for those looking to seriously leverage AI, I believe assembling a strong team is key.

If you're truly committed to developing expert GenAI, start recruiting and training your tech experts, start identifying potential use cases in your business, and start immersing yourself in hands-on experiences now! 🎬

🏎That’s our approach at Koodoo – venturing into the AI terrain and pioneering practical business solutions.

🔥We're passionate about exploring AI's capabilities, seeking to unlock novel business applications and finding innovative ways to elevate human talent through technology. We’re excited to be at the forefront of this journey, shaping the future of AI in business.

✅ Follow Koodoo on LinkedIn today for GenAI insights and real-world applications in financial services.