Founder InternFor the next 2 weeks, I’m documenting my journey as a Prompt Engineering Intern at Virtual Height Pvt Ltd.Real observations about AI workflows, systems, execution, and growth.Daily logs: "sagarithm.in" Day 01 starts tomorrow.

Comments

[Harjot Rana]

Documenting a prompt-engineering internship in public is a great way to learn, and the one thing I'd offer early is that the highest-leverage skill in this role is shifting from prompt engineering to context and system engineering faster than most people do. Crafting clever prompts feels like the job at first, but you hit diminishing returns quickly, and the real reliability comes from what you feed the model (the right context, retrieved and trimmed) and what you build around it (validation that catches bad output, examples that pin the format, a loop that's bounded). Prompts matter, but they're the smallest lever. A few things worth learning hands-on during the two weeks, because they compound: that examples often beat instructions (show, don't tell), that more context isn't better (a tight, relevant context beats a stuffed one), and that you can't improve what you don't measure, so building a tiny eval for your prompts teaches you more than tweaking wording by feel. If you internalize evaluate, don't vibe-check early, you'll be ahead of most people with the title. Treat prompts as one input to a system you measure, not the whole craft. That measure-and-engineer-the-context instinct is core to how I think about Moonshift. As you document the journey, are you planning to set up any eval to judge your prompts objectively, or going mostly on output feel for now?