"Your Ai coach"

A developer built a local, offline internship interview coach powered by Gemma 4B and Ollama for a friend, an HR student preparing for her first internship interviews. The app takes a pasted job offer and optional CV, has the model act as a recruiter asking tailored questions, then returns structured feedback and a score. The developer reports that tightening the output-format prompt, rather than using a larger model, fixed unreliable scoring.

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend https://dev.to/challenges/hacktoberfest-weekend-2026-10-01 An internship interview coach . You paste a job offer and optionally your CV , and the AI plays the recruiter. It asks you questions, listens to your answers, then gives you structured feedback and a score. I built it for Khennel. She stresses about her answers, hesitates, and lacks structure. Worse, she has nobody to practice with who will give her honest feedback. Friends are too nice, and online tools ask for an account, a subscription, or send her answers to someone else's servers. When you're practicing something vulnerable, that last part matters. An internship interview coach that runs 100% locally with Gemma and Ollama Built for my friend Khennel , a Human Resources student preparing her first internship interviews. Built for the DEV Hacktoberfest Weekend Challenge: Build for a Friend Category: Best Use of Gemma . The whole thing runs locally : The flow is simple. The user pastes a job offer and a CV if they want . A first "recruiter" prompt turns the model into an interviewer that asks questions tailored to the offer. Once the answers are in, a second "feedback" prompt analyzes them and returns structured feedback with a score. I learned two things the hard way: A small model sometimes gets the scores wrong. The fix was not a bigger model, it was a tighter prompt: I constrained the output format so the model had less room to wander. Keeping the project small is a feature. One weekend, one clear use case, one complete app. This project only makes sense because the AI is open. It runs locally and offline. Khennel can practice anywhere, even with a bad connection. Her data stays private. Her answers, her CV, her nerves: none of it leaves her laptop. It costs nothing. No subscription, no API bill, no account. A student can use it for free, as often as she wants. The model is swappable. If a better open model shows up tomorrow, I change one line. A closed API would have meant sending a stressed student's CV and unpolished answers to a server, and paying per practice session. For a tool whose whole purpose is "fail safely, then try again", that would have defeated the point. Best Use of Gemma : the whole project is built around Gemma 4B running locally. Thanks for participating