“My big lesson could see that domain technology was more important than the difficulties of algorithmic.”

In a series of spotlight writer, Tds editors interview members of our community in their work in Data Science and Ai, their writing, and their inspiration. Today, we are happy to share our conversations with them Claudia Ng.

Claudia is a businessman for AI and six-year-old data scientists for receiving the production model models in Fintech. He has digged the second and won $ 10,000 in the Web3 Web-Credit Dictionary ML in 2024.

He just won $ 10,000 in the Teaching Court – Congratulations! What is the great lesson you've taken to this program, and how do we make a way that approaches the real-ML world problems?

My big lesson could see that the domain technology was the same as the algorithmic crisis. It was to beat Web3 Seweb3, and even though there was no work with Blockchain or neural networks for credit points, 6+ years gave me a business problem as an accident problem. This idea seems more important than any degree or deep learning technology.

This experience is basically removed from how I come with ML problems in two ways:

First, I learned that the shipment is better than perfection. I spent only 10 hours in the competition and moved the “MVP” method rather than an extremist engineer. This applies directly to the industrial work: Honorable model running in production brings more value than a very configured model in the Jobyter registry.

Second, I have found that many obstacles are mental, not technical. I probably didn't come in because I didn't know web3 or heard as a “competitive” person, but also again. While I was working on using this lesson more, we have changed from checking the chances. I now focus on whether I understand the key issue and that I'm fun, and do you trust that I will be able to find it as I go.

Your work path is destroying business, public policy, machine-study, and ai adviser. What motivated your switch to a mixture of united technology in Ai Freadance World, and what happens at this new chapter? What kind of challenges or clients do you enjoy?

The independent work shift was conducted by wanting to build something I could really know. In the organizational roles, build essential systems give the strength to fall, but you cannot take them or get the excellent credit of their success. Wining the competition showed me that I had the skills causing my solutions rather than I contribute to another person's view. I learned the important skills in the joint ransom, but I am very happy to put them at risks that deeply concern.

I pursue this in two main ways: Reasoning with my scientific science and machine learning technology, and building a product of learning AI. The consultation function provides immediate revenue and stores connected with real business problems, while language product represents my long-term opinion. I learn to build a community and share my journey with my Newsletter newspaper.

As a nine language polyglot, I have deeply thought about the challenges of getting reversal not just the books of the books I am learning another language. I am developing a learner's learner's learner who helps people make real conditions of the world and cultural situations.

What pleases me is the technical challenge of building AI solutions that look at the cultural context and the change of nuantly. On the consultation side, I have the ability to work with companies that seek to solve real problems rather than just using AI because of AI. Even if HIT models work

Many companies yearn to “do something about Ai” but always don't know where to start. What is your normal process to help the new client limit and place forward their first AI system?

I take advantage of – the original way rather than luring ai solutions. Many companies want to “do something about AI” without identifying what a business problem that is trying to solve, which often leads to impressive demeans do not go with a needle.

My normal process follows three steps:

First, I focus on the examination of problems. We identify some pain points about an impossible impact. For example, I have recently worked with the client in the fight against the restaurant. Instead of jumping on “the solution with AI,” we checked customer review data to find patterns. For example, that the information menu driven by complaints, which items are used that generated a positive response, and what apps that are already very often. This data conducted by data has led to more commendation rather than generic Ai.

Second, we explain success before. I insist on the Metrics available as retirement, quality development, or income. If we are unable to measure, we cannot prove it to work. This prevents the greatest intelligence and ensures that we solve real problems, not just to build cool technology.

Third, we walk in practical solutions and conform to the best. Sometimes that is a visual dashboard observation, sometimes it is a rag program, sometimes adding predictive skills. AI does not always replace, but if so, we know exactly why we also use it.

This method has brought positive results. Customers often see the speed of making upgraded decisions and clear data understanding. When I developed my independent habit, I focus on real problems than ai buzzwords has been a key to customer satisfaction and multiplying.

You learned to wish data scientists – What is a common picture you see among people trying to get into the wild, and how do you advise them to avoid them?

I see a big passenger I see trying to read everything instead of focusing on one passage. Many people, including me my own, feel like they need to take every Ai course and are able to get all the ideas before they are “qualified.”

The fact is that data science includes very different roles: from product information scientists using A / B tests on ML Engineergers using models in production. You don't need to be a specialist in everything.

My advice: Choose your Standard first. Find out what role you like most, and focus on that key skills. I have been changed from analyst to ML Gener by learning a deeper learning and taking real projects (you can read my change story here). I include my domain technology in debt and deception risk, and use this to install engineering and business impact.

The key uses these skills in real problems, not putting them in hell. I see this approach always has my Newsletter book and teaching. The people who break them are the first ones who began to build, or they felt ready.

The state of AI roles keeps up. On how new arrivals should they decide where to focus – ML engineering, data analytics, llms, or altogether?

Start with your current set of set and preference, not the most important. I have worked well in different roles (analyst, a data scientist, a ML engineer) and each of which brought important, transmitted skills.

Here's how I can approach the decision:

If you appear in the back of the business: Product Data Sciest Riles is usually a simple login point. Focus on SQL, A / B to view testing data view skills. These roles usually charge an intuition business value over the deep technical skills.

If you have programming information: Consider the ML engineer or ai engineering. The need is high, and you can create existing software development skills.

If you are drawn to infrastructure: MLOS engineering is very demanded, especially additional companies submit ML models and AI models in scale.

The local condition continues to appear, but as mentioned above, domain technology is usually something more than follow the latest trail. I've won that ML race because I understood the dangers of debt, not because I could be very nice algorithms.

Focus on resolving real problems to your domains you understand, and allow technical skills to follow. To learn more about different roles, I write about 5 types of career career science methods here.

What one of AI or science data is the article that thinks most people should write about or one watching trends nearby now?

I am struck with speed and technical quality (TTS) in order to assign real tones to tell. I think most people should write about TTS technology through endangered language.

Like a polyglot who has a loving understanding of cultural understanding, I am glad how AI can help protect the language completely. Most TTS Development focuses on large detailed languages, but there are more than 7,000 languages worldwide, and many are in danger of being annihilated.

It is a pleasure that AI can create a voice compilation of languages that may only have only a few hundred speakers. This is the technology of working for the personality and preservation of their best! When the language die, we lose different ways to think about the earth, certain information systems, and cultural memory fails to be interpreted.

This practice I look closely is the way to learn to read and the word of the Word deal with this technology. We reach a point where you can only need hours than thousands of sound data to create high-quality TTS in new languages, especially using existing multilingual models. While this technology is raising legitimate concerns about misuse, the common use of language shows how we can use these skills by cultural commitment.

As I continue to improve my language learning and build my own consultation. Whether you are in the machine for reading machine or cultural communication tools, the magic happened when it met.

To learn more about Claudia's work and stay in the latest and latest news, you can follow him in TDS, in the last place, or LinkedIn.