Learn to engineer software in the age of AI.

I started using Claude almost two years ago. At first, I found it fascinating, but later, while building real AI projects, it started to feel like an actual thinking partner.

One project I probably wouldn't have attempted on my own was an accessibility browser extension during my AI engineering co-op, designed to help people with accessibility needs. Claude changed my approach when the project felt too big and messy: instead of trying to solve everything at once, I gave it the full context — the goal, the constraints, what I had tried, and what confused me — and used its perspective to compare options, think through the tech stack, and break the work into smaller steps.

It didn't replace my thinking; it helped me think more clearly. More than anything, having access to it made me more confident as a learner and a builder.

Coming into AI 110 as a student, I found Claude made a big difference in how I approached learning and building. One project where it really helped me was my Music Discovery Agent. Before using Claude, I probably would have kept the project much smaller or avoided some of the harder parts.

What stood out to me most was that Claude did not just give me answers. It helped me ask better questions. When I was working through confidence scoring and recommendation logic, Claude gave me ideas that helped me think more clearly — while also pushing me to review its suggestions critically.

Now that I am back as a TF, I notice students using Claude in a similar way. The biggest difference is that students are more willing to start. Having access to Claude made AI 110 feel more approachable and more creative. It gave me the confidence to build something more ambitious, and now I get to help students learn how to use it responsibly and effectively, too.

I grew up in Zambia and went to India to study a Bachelor of Pharmacy. Somewhere during pharmacy school, I started teaching myself to code — building small tools on the side because I kept seeing problems software could solve and couldn't let them go. Pharmacist by training, self-taught full-stack developer by obsession.

I've been using Claude for about two years now, and it's quietly become how I build. Early on, it helped me get unstuck — explaining a concept I'd never been taught, debugging something at 2am, turning an idea in my head into working code faster than I could alone. With no formal CS background, having something that could meet me at my level and pull me up a notch was the difference between abandoning a project and shipping it.

Last year I moved to New York for a master's in Biotechnology and Entrepreneurship at Yeshiva University, and I knew I wanted to build AI products properly — not just dabble. That's what brought me to CodePath. I took AI110 in the spring and came back for AI201 this summer, and that's when what I used Claude for really changed. I went from "make it work" to "make it trustworthy."

I'm building RegComplyAI, a tool that helps biotech founders navigate the FDA approval process — cutting cost and time, reducing paperwork, and raising quality. My first version looked good, but nobody trusted it: confident answers, no citations, no way to verify. Working through it with Claude, I stopped asking "does it work?" and started asking "how does someone who isn't me know it's right — and what happens when it's wrong?" I rebuilt the whole thing around that question: guardrails that catch bad inputs, retrieval with citations so every answer points to its source, and an agentic critic that reviews outputs before a human ever sees them.

That's the real shift. Claude didn't just help me write code — it helped me learn how production AI is actually built. The questions I ask before shipping anything are different now, and that's mine to keep.