Within the Writer Highlight collection, TDS Editors chat with members of our group about their profession path in information science and AI, their writing, and their sources of inspiration. At present, we’re thrilled to share our dialog with Claudia Ng.
Claudia is an AI entrepreneur and information scientist with 6+ years of expertise constructing manufacturing machine studying fashions in FinTech. She positioned second and gained $10,000 in a Web3 credit score scoring ML competitors in 2024.
You recently won $10,000 in a machine learning competition — congratulations! What was the largest lesson you took away from that have, and the way has it formed your method to real-world ML issues?
My largest lesson was realizing that area experience issues greater than algorithmic complexity. It was a Web3 credit score scoring ML competitors, and regardless of by no means having labored with blockchain information or neural networks for credit score scoring, my 6+ years in FinTech gave me the enterprise instinct to deal with this as an ordinary credit score threat drawback. This attitude proved extra helpful than any diploma or deep studying specialization.
This expertise essentially shifted how I method ML issues in two methods:
First, I realized that shipped is best than good. I spent solely 10 hours on the competitors and submitted an “MVP” method reasonably than over-engineering it. This is applicable on to trade work: an honest mannequin operating in manufacturing delivers extra worth than a extremely optimized mannequin sitting in a Jupyter pocket book.
Second, I found that almost all boundaries are psychological, not technical. I nearly didn’t enter as a result of I didn’t know Web3 or really feel like a “competitors particular person”, however on reflection, I used to be overthinking it. Whereas I’m nonetheless engaged on making use of this lesson extra broadly, it has modified how I consider alternatives. I now deal with whether or not I perceive the core drawback and whether or not it excites me, and belief that I’ll be capable to determine it out as I am going.
Your profession path spans enterprise, public coverage, machine studying, and now AI Guide. What motivated your shift from company tech to the AI freelance world, and what excites you most about this new chapter? What sorts of challenges or purchasers are you most excited to work with?
The shift to unbiased work was pushed by wanting to construct one thing I might actually personal and develop. In company roles, you construct helpful methods that outlive your tenure, however you may’t take them with you or get ongoing credit score for his or her success. Profitable this competitors confirmed me I had the abilities to create my very own options reasonably than simply contributing to another person’s imaginative and prescient. I realized helpful abilities in company roles, however I’m excited to use them to challenges I care deeply about.
I’m pursuing this via two primary paths: consulting initiatives that leverage my information science and machine studying experience, and constructing an AI language studying product. The consulting work offers speedy income and retains me related to actual enterprise issues, whereas the language product represents my long-term imaginative and prescient. I’m studying to construct in public and sharing my journey via my newsletter.
As a polyglot who speaks 9 languages, I’ve thought deeply in regards to the challenges of attaining conversational fluency and never simply textbook data when studying a overseas language. I’m growing an AI language studying accomplice that helps folks follow real-world eventualities and cultural contexts.
What excites me most is the technical problem of constructing AI options that take note of cultural context and conversational nuance. On the consulting facet, I’m energized by working with firms that wish to remedy actual issues reasonably than simply implementing AI for the sake of getting AI. Whether or not it’s engaged on threat fashions or streamlining data retrieval, I really like initiatives the place area experience and sensible AI intersect.
Many firms are desirous to “do one thing with AI” however don’t at all times know the place to start out. What’s your typical course of for serving to a brand new shopper scope and prioritize their first AI initiative?
I take a problem-first method reasonably than lead with AI options. Too many firms wish to “do one thing with AI” with out figuring out what particular enterprise drawback they’re making an attempt to unravel, which often results in spectacular demos that don’t transfer the needle.
My typical course of follows three steps:
First, I deal with drawback prognosis. We determine particular ache factors with measurable affect. For instance, I lately labored with a shopper within the restaurant area going through slowing income development. As a substitute of leaping to an “AI-powered answer,” we examined buyer overview information to determine patterns. For instance, which menu gadgets drove complaints, what service components generated constructive suggestions, and which operational points appeared most continuously. This data-driven prognosis led to particular suggestions reasonably than generic AI implementations.
Second, we outline success upfront. I insist on quantifiable metrics like time financial savings, high quality enhancements, or income will increase. If we will’t measure it, we will’t show it labored. This prevents scope creep and ensures we’re fixing actual issues, not simply constructing cool expertise.
Third, we undergo viable options and align on the perfect one. Typically that’s a visualization dashboard, typically it’s a RAG system, typically it’s including predictive capabilities. AI isn’t at all times the reply, however when it’s, we all know precisely why we’re utilizing it and what success appears to be like like.
This method has delivered constructive outcomes. Shoppers sometimes see improved decision-making pace and clearer information insights. Whereas I’m constructing my unbiased follow, specializing in actual issues reasonably than AI buzzwords has been key to shopper satisfaction and repeat engagements.
You’ve mentored aspiring information scientists — what’s one frequent pitfall you see amongst folks making an attempt to interrupt into the sphere, and the way do you advise them to keep away from it?
The largest pitfall I see is making an attempt to study all the things as a substitute of specializing in one function. Many individuals, together with myself early on, really feel like they should take each AI course and grasp each idea earlier than they’re “certified.”
The fact is that information science encompasses very totally different roles: from product information scientists operating A/B assessments to ML engineers deploying fashions in manufacturing. You don’t should be an knowledgeable at all the things.
My recommendation: Decide your lane first. Determine which function excites you most, then deal with sharpening these core abilities. I personally transitioned from analyst to ML engineer by intensely learning machine studying and taking over actual initiatives (you may learn my transition story here). I leveraged my area experience in credit score and fraud threat, and utilized this to characteristic engineering and enterprise affect calculations.
The secret’s making use of these abilities to actual issues, not getting caught in tutorial hell. I see this sample always via my e-newsletter and mentoring. Individuals who break via are those who begin constructing, even once they don’t really feel prepared.
The panorama of AI roles retains evolving. How ought to newcomers determine the place to focus — ML engineering, information analytics, LLMs, or one thing else completely?
Begin along with your present talent set and what pursuits you, not what sounds most prestigious. I’ve labored throughout totally different roles (analyst, information scientist, ML engineer) and every introduced helpful, transferable abilities.
Right here’s how I’d method the choice:
When you’re coming from a enterprise background: Product information scientist roles are sometimes the best entry level. Deal with SQL, A/B testing, and information visualization abilities. These roles typically worth enterprise instinct over deep technical abilities.
You probably have programming expertise: Take into account ML engineering or AI engineering. The demand is excessive, and you’ll construct on current software program improvement abilities.
When you’re drawn to infrastructure: MLOps engineering is very in demand, particularly as extra firms deploy ML and AI fashions at scale.
The panorama retains evolving, however as talked about above, area experience typically issues greater than following the most recent development. I gained that ML competitors as a result of I understood credit score threat fundamentals, not as a result of I knew the fanciest algorithms.
Deal with fixing actual issues in domains you perceive, then let the technical abilities observe. To study extra about totally different roles, I’ve written in regards to the 5 forms of information science profession paths here.
What’s one AI or information science matter you assume extra folks ought to be writing about or one development you’re watching carefully proper now?
I’ve been blown away by the pace and high quality of text-to-speech (TTS) expertise in mimicking actual conversational patterns and tone. I believe extra folks ought to be writing about TTS expertise for endangered language preservation.
As a polyglot who’s keen about cross-cultural understanding, I’m fascinated by how AI might assist stop languages from disappearing completely. Most TTS improvement focuses on main languages with large datasets, however there are over 7,000 languages worldwide, and plenty of are vulnerable to extinction.
What excites me is the potential for AI to create voice synthesis for languages that may solely have a couple of hundred audio system left. That is expertise serving humanity and cultural preservation at its finest! When a language dies, we lose distinctive methods of serious about the world, particular data methods, and cultural reminiscence that may’t be translated.
The development I’m watching carefully is how switch studying and voice cloning are making this technically possible. We’re reaching some extent the place you would possibly solely want hours reasonably than 1000’s of hours of audio information to create high quality TTS for brand spanking new languages, particularly utilizing current multilingual fashions. Whereas this expertise raises legitimate issues about misuse, functions like language preservation present how we will use these capabilities responsibly for cultural good.
As I proceed growing my language studying product and constructing my consulting follow, I’m always reminded that essentially the most attention-grabbing AI functions typically come from combining technical capabilities with deep area understanding. Whether or not it’s constructing machine studying fashions or cultural communication instruments, the magic occurs on the intersection.
To study extra about Claudia‘s work and keep up-to-date along with her newest articles, you may observe her on TDS, Substack, or Linkedin.
